Session G5: Focus Session: Computational Discovery and Design of New Materials: Thermodynamics and Mechanical Properties

Materials Design based on Predictive Ab Initio Thermodynamics

A key requirement in developing predictive multi-scale modeling is the availability of accurate computational tools determining energies not only at T = 0 K but also under realistic conditions, i.e., at finite temperature. Combining accurate first principles calculations with mesoscopic/macroscopic thermodynamic and/or kinetic concepts allows now to address this issue and to determine free energies and derived thermodynamic quantities such as heat capacity, thermal expansion coefficients, and elastic constants with an accuracy that matches and often even rivals available experimental data. In the talk a brief overview of the fundamentals and recent developments of combining modern fully parameter-free ab initio methods with thermodynamic concepts will be given with special emphasize on structural materials. The flexibility and the predictive power of these approaches and the impact they can have in developing new strategies in materials design will be discussed e.g. for modern high strength TWIP/TRIP steels, for understanding failure mechanisms such as hydrogen embrittlement, or for identifying chemical trends in the performance of light weight metallic alloys. Work has been done in collaboration with Fritz Kormann, Blazej Grabowski, and Tilmann Hickel.   

Metastable Al-rich phases in the Al-Sm system: A genetic-algorithm study

Metallic glasses formed by Al and about 10{\%} rare earths such as Sm are important high-strength-low-density materials. Various metastable crystalline phases are formed in the early stages of the devitrification of Al90Sm10 glasses. Identification of these phases is crucial to understand the phase selection during amorphization and devitrification processes, and thus provides critical information for the control of microstructures in order to obtain desired mechanical properties. In this study, we use a genetic algorithm to systematically study the low energy Al-rich phases of the Al-Sm system. We discovered a new Al5Sm phase that matches excellently with the experimentally detected M1 phase in lattice parameters as well as diffraction patterns. In addition, we established the energy landscape as a function of Al composition on the Al-rich side of the phase diagram, and found key geometries of Sm-centered local clusters which could serve as building blocks for other metastable phases.   

Prediction of previously unreported 18-electron ABC materials via first-principles thermodynamics

The eighteen electron s2p6d10 ABC compounds derived from an s2p6 binary plus a column X element such as Ni,Pd,Pt have recently been proposed as new topological insulators and thermoelectric materials. Yet, many potentially stable 18 electron compounds are not reported in standard compilations, raising the question if they are stable or unstable. Here we use ``first-principles thermodynamics'' [1] to evaluate the thermodynamic stability of the 401 currently undocumented s2p6d10 ABC materials in the groups I-X-VII, II-X-VI, III-X-V, IV-X-IV, II-IX-VII, III-IX VI, IV-IX-V, and V-IX-IV (but excluding the elements Cs, B, Tl, and Po). The calculation follows three steps: (1) Establishing the lowest energy structure of the ternary. (2) Calculating the energies of documented and undocumented potentially competing phases in the A-B-C system. (3) Examining the dynamic stability of the new compound. We report the stable structures, formation enthalpies (corrected for DFT errors) of the new stable compounds, and document the competing phases of the predicted unstable compounds. Recently one of the predicted stable compounds--TaCoSn--has been successfully synthesized. [1] X. Zhang, L. Yu, A. Zakutayev, and A. Zunger, Adv. Funct. Mater. 22, 1425, (2012).   

Computational Study of the Phase Diagram of Tungsten-Nitride

There has been considerable interest in the Tungsten Nitride (WN) system, as various calculations predict that structures of WN$_2$, WN$_3$, and WN$_4$ may be ultra-compressible and perhaps as hard as diamond. The predicted stability of these structures is based on vibrational stability and a negative formation enthalpy ($\Delta$H) relative to $\alpha$N$_2$ and bcc W. A negative value of $\Delta$H does not guarantee a compound exists, as it may be above the tie-line constructed from other compounds with lower enthalpies. The determination of the tie-line is complicated by the fact that the ground state of WN is not well determined by experiment or theory. We have used AFLOW, extended to include a large variety of Nitrogen containing structures, to compute the formation energy of the WN system over a large range of compositions. In this talk we discuss the ground state of the WN system, the relationship of our predicted structure to experiment, and the relative stability of the possible ``superhard'' WN$_x$ compounds. We find that this behavior depends on whether a GGA or LDA functional is used, probably because of the inability of these functionals to handle van der Waals forces in N$_2$ crystals.   

Elastic Constants and Phonons of Tungsten-Nitride from First Principles

Certain Tungsten Nitride (WN) crystal structures have been found to exhibit tendencies for exceptional hardness. Some researchers [S. Aydin et al., J. Mater. Res. 27, 1705 (2012)] have made the claim that these structures have hardness qualities that rival diamond. There are three specific structures with unique compositions that are of interest. By calculating the bulk and shear moduli as well as analyzing phonon dispersion plots, the properties of these structures can be compared to known structures like diamond. We used VASP density-functional methods implemented within the MedeA software package to strain each structure in a series of directions in increasing amounts. A simple linear fit of stress vs. strain found that the leading structure in terms of thermodynamic stability has elastic constants of C$_{11}$ = 753 GPa, C$_{12}$ = 126 GPa, and C$_{44}$ = 172 GPa. These constants, while high, are significantly lower than diamond's. This indicates that previous calculations may have been mistaken in predicting the qualities of the WN system. Some of the difference between our results is due to the exchange-correlation functional chosen, namely, LDA vs. GGA.   

Building Symmetry During Crystal Structure Prediction

Unconstrained crystal structure prediction is difficult in large cells since the number of free variables increases rapidly with the number of atoms that are included. We describe a method that builds symmetry on the fly during crystal structure prediction and uses this symmetry to reduce the dimensionality of the search space. We apply this method to Monte Carlo-based crystal structure prediction and show that simulations that build symmetry greatly outperform those that do not, both in average and fastest times to find the ground state structure.   

Simulations of the Structure and Properties of Large Icosahedral Boron Clusters Based on a Novel Semi-Empirical Hamiltonian

A successful development of a parameterized semi-empirical Hamiltonian (SCED-LCAO) for boron based on a LCAO framework using a sp$^{3}$ basis set will be discussed. The semi-empirical Hamiltonian contains environment-dependency and electron screening effects of a many-body Hamiltonian and allows for charge self-consistency. We have optimized the parameters of the SCED-LCAO Hamiltonian for boron by fitting the properties (e.g., the binding energy, bond length, etc.) of boron sheets, small clusters and boron alpha to first-principles calculations based on DFT calculations. Although extended phases of boron alpha and beta have been studied, large clusters of boron with icosahedral structures such as those cut from boron alpha are difficult if not impossible to simulate with ab initio methods. We will demonstrate the effectiveness of the SCED-LCAO Hamiltonian in studying icosahedral boron clusters containing up to 800 atoms and will report on some novel boron clusters and computational speed. [1] C. Leahy, et al, Phys. Rev. B 74,155408 (2006). [2] P. Tandy, et al, Bulletin of the APS, 2009 APS March Meeting Vol. 54, Num.1, Sess. D26, [3] Ming Yu, et al, J. Chem. Phys. 130,184708 (2009). [4] Ming Yu, S.Y. Wu, and C.S. Jayanthi, Physica E 42, 1 (2009).   

The Structure, Stability, and Properties of a One-Dimensional $\alpha $-Boron Structure

Boron is an electron deficient element exhibiting a complex and versatile chemistry. In this work we have performed a preliminary study on the structural stability and electronic properties of one-dimensional $\alpha $-boron structures based on the SCED-LCAO molecular dynamics scheme (MD) [PRB 74, 15540 (2006)]. The one-dimensional $\alpha $-boron structures were generated by constructing icosahedra B$_{12}$ clusters, referred as $\alpha $-boron balls, and arranging them in one-dimension. Such structures were stabilized through the simulated annealing based on the SCED-LCAO MD. We found that: (1) the $\alpha $-boron ball is compressed in comparison to its bulk counterpart ($\alpha $-phase); (2) the distance between ``$\alpha $-boron balls'' is shorter in the center of the chain than that at the two ends and decreases as the length of the chain increases; (3) the HOMO-LUMO gap is very small ($\sim$1 meV) in the finite chains, but it opens up when the chain length becomes infinite. The optimized lattice constant of the infinite $\alpha $-boron chain was found to be 2.998 {\AA} and its energy gap is found to be 0.74 e. The stability and properties of ring-shaped one-dimensional $\alpha $-boron structures will also be discussed.   

A framework for the design of nanomaterials with targeted applications

Engineered nanomaterials are the key building blocks of modern optoelectronic devices and understanding their structure-property relationship can lead to breakthroughs in device design. Even if we suspect that there may be many structures that are consistent with the targeted property, finding a single one may be a daunting, but often also unnecessary, task because of the enormous space of material parameters. Here, I will show how to parametrize the structure via a set of structural features using the targeted physical property as a constraint. Structural features are extracted from our procedure such that they are relevant to the targeted property/ies and linked to the underlying full atomistic structure. I will discuss the conditions under which the representation of the structure via structural features can be very similar, ideally equivalent, to the full structure \emph{only} relative to our targeted properties. Finally, I will demonstrate the importance and validity of the approach using the example application of engineered nanomaterials for third-generation photovoltaics solar cells.   

Optimizing the design of artificial lattices

Artificial crystal lattices are a powerful tool for studying other condensed matter systems because they are easily tunable and may be controllably fabricated in the laboratory. For example, artificial graphene can be created out of arrays of CO molecules arranged on a Cu(111) surface. We generalize the idea of artificial graphene, and propose new, unusual band structures that can result from different types of artificial lattices. Because of the high degree of freedom in creating artificial lattices, the task of systematically designing artificial lattices that exhibit these unusual band structures is non-trivial. We address this optimization problem, and show that new physics can be observed in presently feasible artificial crystal lattices. This work was supported by NSF grant No. DMR10-1006184 and U.S. DOE under Contract No. DE-AC02-05CH11231. Computational resources have been provided by NERSC and XSEDE.   

Modeling fracture of random media via stochastic molecular mechanics

Inspired by recent experimental results suggesting that the heterogeneous distribution of the elastic modulus in bone tissue leads to increased toughness, we determine the toughness modulus of a flawed discrete particle system with stochastic elastic properties. We consider an elastic solid in plane strain conditions in uniaxial tension with a Young's modulus distribution modeled as a 2-d Gaussian process with covariance modeled as an exponential kernel. We solve the problem from a continuum perspective, both employing spectral methods with stochastic finite elements and Monte Carlo methods with conventional finite elements. We also analyze an equivalent discrete particle system, modeled as a spring bead network of FCC-lattices. Our results validate the persistence of the Cauchy Born rule in a stochastic system. We then analyze a flawed discrete particle system to assess the effect of heterogeneity on fracture properties. By studying the fracture mechanics of this system with a range of variance and correlation length parameters in the exponential kernel we gain fundamental insights in to the essential length scales of heterogeneity critical to enhanced fracture properties. This validated stochastic molecular mechanics framework further supports the inverse computation of local elastic properties, not accessible with continuum mechanics, to tailor global mechanical properties such as the fracture toughness. Specifically, Markov Chain Monte Carlo can be used to infer the elastic and geometric parameters. Our work sets the foundation for stochastic modeling in a micromechanical environment and unveils mechanisms by which mechanical behavior can be tailored due to increasingly heterogeneous mechanical properties.   

The unsuspected origin of gold's nobleness

Understanding the ``inertness'' of Au toward oxidizing agents - appreciated since long before the beginning of recorded history -- has remained a challenge. Its nobleness has long been attached to its weak interaction with adsorbates, which contrasts with the fact that Au forms stable alloys and can be made reactive. Density-functional-theory (DFT) calculations of the binding energy (BE) of O on (111) surfaces, in fact, have shown that Au stands out for rendering the weakest BE. Here, we reveal the origin of gold's unique inertness by revising the adsorption of this prototype oxidizing agent on several (111) metal surfaces. We show via DFT that, judging by BE of O on Au(111) and Ag(111), e.g., \textit{both} the d-band-center argument \textit{and} analysis of the electronic density of states fail to describe the relatively low reactivity of Au. Nevertheless, we establish that, rather than failure of the above paradigms, a key element to understand BE of adsorbates has been left behind so far. Namely, we demonstrate that, although BE of O is higher on Ag(111) than on Au(111), (1) The local Au-O bonds are \textit{indeed stronger} than the Ag-O ones; (2) the low BE of O on Au is, paradoxically, caused by an unusually large perturbation on Au-Au bonds upon O adsorption.