Session A7: Focus Session: Computational Design of Materials - Structure Prediction

Crystal structure prediction: a novel approach based on minima hopping

With increasing computational resources the prediction of crystal structures from first principle calculations has become feasible, but still remains a demanding task. A reliable method to perform an efficient, systematic search for the ground state structure based solely on the system's composition is essential. Motivated by the promising results of the minima hopping method obtained on isolated systems, we have generalized the algorithm for crystal structure prediction. Optimized moves in the configurational space spanned by both atomic coordinates and simulation cell variables are performed to escape from local enthalpy minima, and revisiting known minima is avoided, thus allowing a fast exploration of the enthalpy surface. The predictive power of the novel method has been shown in several applications, of which the following will be presented. Superconducting phases in hydrogen rich materials were investigated, leading to the discovery of novel ground state structures. For the longstanding question of the crystal structure of cold compressed graphite a new candidate phase could be identified to perfectly match experimental results. And at last, new low energy structures for materials with possible applications in hydrogen storage are presented.   

Adaptive Genetic Algorithm method for Crystal Structure Prediction and Materials Discovery

We developed a fast and efficient method for crystal structure prediction and materials discovery. The method is based on the cut-and-paste genetic algorithm (GA) scheme introduced by Deaven and Ho [1]. In the evaluation of energies of target structures, first-principles calculations are accurate but time-consuming. Our method performs GA searches uses auxiliary model potentials to screen the energy of candidate structures, selecting only a few for more extensive first principles evaluation. Parameters of the auxiliary potentials are adaptively adjusted to reproduce the first-principles results during the course of the GA search. Our method combines the speed of empirical potential searches with the accuracy of first principles calculations. We will present results on applications to various systems including metallic alloys and ultrahigh pressure SiO$_{2}$, H$_{2}$O and Mg-Si-O systems. \\[4pt] [1] D. M. Deaven, K. M. Ho, Phys. Rev. Lett. 75, 288 (1995).   

Periodic structure optimization via local heat pulse-quench cycles employing the GULP code

We present an optimization algorithm for problems with many continuous degrees of freedom and a huge number of local minima. It is based on the thermal cycling approach, originally developed for combinatorial optimization tasks [1]. The main idea is to cyclically disturb a few randomly chosen degrees of freedom of the current best local minimum and to quench this state by a highly efficient local search code. As the optimization proceeds, the amplitude of the disturbance slowly decreases. This approach is applied to a lattice structure prediction problem. We use the general utility lattice program (GULP) by J.D. Gale and co-workers [2] for local search. As test, the hypothetical periodic Mg$_{10}$Al$_4$Ge$_2$Si$_8$O$_{36}$ compound is studied, where both the cell parameters and the atom positions are free to vary. The results demonstrate that the proposed procedure is robust and far more efficient than the previous approaches to this problem by means of multi-start local search, simulated annealing, and evolutionary algorithms in Ref. 3. \\[4pt] [1] A. M\"{o}bius et al., Phys. Rev. Lett. 79 (1997) 4297. \\[0pt] [2] J.D. Gale and A.L Rohl, Mol. Simul. 29 (2003) 291. \\[0pt] [3] A.R. Oganov et al., in ``Modern Methods of Crystal Structure Prediction,'' ed. A.R. Organov, (Wiley, 2011), p. 223.   

Mapping the Materials Genome through Combinatorial Informatics

The recently announced White House Materials Genome Initiative provides an exciting challenge to the materials science community. To meet that challenge one needs to address a critical question, namely \textit{what} is the materials genome? Some guide on how to the answer this question can be gained by recognizing that a ``gene'' is a carrier of information. In the biological sciences, discovering how to manipulate these genes has generated exciting discoveries in fundamental molecular biology as well as significant advances in biotechnology. Scaling that up to molecular, cellular length scales and beyond, has spawned from genomics, fields such as proteomics, metabolomics and essentially systems biology. The ``omics'' approach requires that one needs to discover and track these ``carriers of information'' and then correlate that information to predict behavior. A similar challenge lies in materials science, where there is a diverse array of modalities of materials ``discovery'' ranging from new materials chemistries and molecular arrangements with novel properties, to the development and design of new micro- and mesoscale structures. Hence to meaningfully adapt the spirit of ``genomics'' style research in materials science, we need to first identify and map the ``genes'' across different materials science applications On the experimental side, combinatorial experiments have opened a new approach to generate data in a high throughput manner, but without a clear way to link that to models, the full value of that data is not realized. Hence along with experimental and computational materials science, we need to add a ``third leg'' to our toolkit to make the ``Materials Genome'' a reality, the science of Materials Informatics. In this presentation we provide an overview of how information science coupled to materials science can in fact achieve the goal of mapping the ``Materials Genome''.   

Diamondlike carbo-boride C5B compound? Testing crystal structures and stability with Global Space Group Optimization ( GSGO)

Diamond-like C5B has been alleged to be a new exciting material discovered at high pressures and temperatures. Using density-functional based evolutionary Global Space Group Optimization (GSGO) we established the likely structures of BC3 and BC5 phases and the known B4C compound. Examining the ground state line between solid Carbon and solid Boron we find only B4C is a ground state structure. C5B and C3B are high energy structure at high pressure. The reaction between BC3 and solid carbon producing C5B is found to have a positive reaction enthalpy that depends only weakly on pressure. In contrast to the previous reports, we argue that at 0 T the BC5 is less stable than BC3 in access of carbon. However, a small positive enthalpy of the reaction BC5 $\rightarrow$ BC3 +2C cannot rule out formation of the BC5 from the BC3 precursor material as thermally stabilized due to vibrational entropy or/and disorder effects.   

Hands-free thermodynamic alloy modeling of $\sim$700 binary alloys using a Bayesian approach: Part I

Bayesian approaches have become useful in recent years as increasing computing power has made them practical. Bayes rule itself is nothing more than a simple statement of conditional probability but can be used to make strong inferences. We discuss the general idea behind Bayes rule and how to use it to build physical models. Using a database of about 150,000 first principle calculations, we are building models for $\sim$700 binary alloys.   

Hands-free thermodynamic alloy modeling of $\sim$700 binary alloys using a Bayesian approach: Part II

In recent years, Bayesian statistics has become more popular as a scientific tool. This is mainly due to advances in computing power, which make the Bayesian formalism tractable. Baye's rule, which is a simple statement of conditional probability, is the foundation for Bayesian statistics. When used in conjunction with sampling algorithms, such as Metropolis-Hastings and Gibbs, Baye's rule provides a powerful framework for characterizing important parameters and for making inference. We demonstrate how the Bayesian framework can be applied to alloy modeling, providing the solution to several typical problem areas in this field. Together with our large database of first-principles data this Bayesian framework helps us quickly and accurately characterize hundreds of binary alloys.   

Sorting stable from unstable hypothetical compounds and determining the electronic structure of interesting candidates : The case of Half-Heusler Filled Tetrahedral ABX structures

Electronic structure theory has recently been used to propose hypothetical compounds, seeking new useful functional materials. In some cases, such hypothetical compounds in presumed crystal structures may be significantly higher in energy than (i) their lowest-energy structures or, than (ii) a combination of their constituents. We use the first-principles thermodynamics to sort (i) the lowest-energy structure and (ii) the thermodynamic stability with respect to disproportionation of the 488 unreported ABX octet compounds. We find that as many as 235 of the 488 are unstable with respect to decomposition, whereas other 235 of the unreported compounds are predicted to be thermodynamically stable. (18 materials are too close to call). The electronic structures of these predicted stable compounds are evaluated based on GW approximation for electron's self-energy. To support the theoretical predictions of new materials, thin film samples of AgYGe were synthesized using combinatorial Pulsed Laser Deposition. AgYSi-type ground state crystal structure and Y-rich composition stability range of AgYGe agree well with theory.   

Band-structure predictions for A2BX4 discovery compounds

The inverse design of materials requires to predict the existence and the properties of previously unknown materials. We have performed a computational search for thermodynamically stable materials within the family of A2BX4 compounds (A, B = main group and 3d cations; X = O, S, Se, Te) resulting in the theoretical discovery of about 100 previously unreported compounds. The challenge for the prediction of band-structures and optical spectra is to obtain accurate results for a wide range of materials within a single computational scheme, so that unknown materials can be predicted with confidence. Whereas the main group chalcogenides are rather accurately predicted by many-body GW calculations, large deviations from experiment are observed for many 3d oxides. In particular, we find that the 3d orbitals consistently occur at too high energies, independent on whether they are occupied (e.g., Cu2O) or unoccupied (e.g., TiO2). While the exact nature of these issues are under investigation, we pursue here a pragmatic approach, using attractive on-site potentials with a single parameter for each 3d element, which leads to good agreement with experiment for binary and ternary 3d oxides. We use this approach to predict the band-structures of the discovery compounds.   

New Crystal Structures Identified for PtO and PtO$_2$ using Density Functional Theory Calculations

Platinum plays an important role in catalysis and electrochemistry, and it has been known that the direct interaction of oxygen with Pt surfaces can lead to the formation of platinum oxides (PtO$_x$), which can affects the reactivity. To contribute to the atomistic understanding of the atomic structure of PtO$_x$, we report a density functional theory study of the atomic structure of bulk PtO$_x$ ($1 \leq x \leq 2$). From our calculations, we identified a lowest energy structure (GeS-type, space group $Pnma$) for PtO, which is 0.181~eV lower in energy than the structure suggested by Moore and Pauling (PtS-type). Furthermore, two atomic structures were identified for PtO$_2$, which are almost degenerate in energy with the lowest energy structure reported so far for PtO$_2$ (CaCl$_2$-type). Based on our results and analysis, we suggest that Pt and O atoms tends to form octahedron motifs in PtO$_x$ even at lower O composition by the formation of Pt$-$Pt bonds.   

Simulation of structural and electronic properties of amorphous tungsten oxycarbides

Electron beam induced deposition with tungsten hexacarbonyl W(CO)$_6$ as precursors leads to granular deposits with varying compositions of tungsten, carbon and oxygen. Depending on the deposition conditions, the deposits are insulating or metallic. We employ an evolutionary algorithm to predict the crystal structures starting from a series of chemical compositions that were determined experimentally. We show that this method leads to better structures than structural relaxation based on guessed initial structures. We approximate the expected amorphous structures by reasonably large unit cells that can accommodate local structural environments that resemble the true amorphous structure. Our predicted structures show an insulator to metal transition close to the experimental composition at which this transition is actually observed. Our predicted structures also allow comparison to experimental electron diffraction patterns.   

Predicting Two-Dimensional Boron-Carbon Compounds by the Global Optimization Method

We adopt a global optimization method to predict two-dimensional (2D) nanostructures through the particle-swarm optimization (PSO) algorithm. By performing PSO simulations, we predict new stable structures of 2D boron-carbon (B-C) compounds for a wide range of boron concentrations. Our calculations show that: (1) All 2D B-C compounds are metallic except for BC3 which is a magic case where the isolation of carbon six-membered ring by boron atoms results in a semi-conducting behavior. (2) For C-rich B-C compounds, the most stable 2D structures can be viewed as boron doped graphene structures, where boron atoms typically form 1D zigzag chains except for BC3 in which boron atoms are uniformly distributed. (3) The most stable 2D structure of BC has alternative carbon and boron ribbons with strong in-between B-C bonds, which possesses a high thermal stability above 2000 K.   

Quantum confined Schottky barriers: Tuning the Schottky-Mott and Bardeen Limits

Even though metal-semiconductor junctions are an essential component in electronic devices, an atomistic understanding of the electronic structure of such junctions has remained elusive. Hundreds, if not thousands, of atoms may be required to model such interfaces, owing to lattice mismatches at the interface. The absence of a detailed understanding of the interface structure has inhibited the application of electronic structure calculations to examine the evolution of Fermi level pinning. Here we capitalize on recent computational breakthroughs and apply them to a prototypical system: a Pb(111) film on a Si(111) substrate. We consider up to 1,500 atoms in our first principle calculations based on real space pseudopotentials, and explicitly model the atomistic details of the metal-semiconductor interface. We propose a pinning mechanism for the Fermi level that depends critically on the role of quantum confinement in the overlaying metallic film. By changing the thickness of the overlying film, the pinning mechanism can be tuned from the abrupt interface description of Schottky and Mott to the metal induced gap state description of Bardeen.