Accelerated molecular dynamics methods offer a powerful way to push atomistic simulations beyond the microsecond time scale that molecular dynamics is limited to in its standard implementation. These methods are designed for infrequent-event systems, and their conceptual basis is that the time between events is shortened by coaxing the trajectory into finding the next event more quickly, but in some way that does not alter the relative probabilities of the different possibile events. When energy barriers are high, these methods, hyperdynamics, parallel-replica dynamics, and temperature accelerated dynamics, can achieve substantial speedups over direct molecular dynamics, sometimes many orders of magnitude. However, all three of the methods suffer from an intrinsic scaling with system size that makes them much less efficient for large systems, e.g., more than ~1000 atoms. In the case of hyperdynamics, acceleration is achieved by running dynamics on a potential surface that is augmented by a bias potential that raises the potential in the basins. A key requirement is that this bias potential be zero at any dividing surface between states. As the system size is increased, the probability increases that there is a distortion somewhere that takes the system close to a dividing surface, and hence turns off the bias potential. This guarantees that for large enough systems, no matter what form of bias potential is employed, the boost factor in hyperdynamics falls to unity (i.e., no speedup). In this talk, I will describe a new formulation of hyperdynamics in which the bias is applied locally, rather than globally. The method and its implementation involve some subtleties that don't arise in standard hyperdynamics, but the result is a method with a very weak scaling with system size, so that it can be applied to extremely large systems.

As the available computational capacity for scientific computing is growing first principles Born-Oppenheimer molecular dynamics is becoming an increasingly important tool for a wide range of problems in materials science, chemistry and molecular biology. Born-Oppenheimer molecular dynamics based on density functional theory offers a very accurate quantum mechanical approach to atomistic simulations that is more reliable and general compared to classical molecular dynamics. Unfortunately, Born-Oppenheimer molecular dynamics simulations are often limited by a very high computational cost or by fundamental problems such as unbalanced phase space trajectories, numerical instabilities and a systematic long-term energy drift. These problems become particularly severe in combination with reduced complexity or linear scaling algorithms that are necessary for the study of large systems. We have recently taken some steps toward a new generation of first principles molecular dynamics, which combines some of the best features of regular Born-Oppenheimer molecular dynamics and Car-Parrinello molecular dynamics, while avoiding their most serious shortcomings. The new dynamics is given in terms of an extended Lagrangian framework, where auxiliary extended electronic degrees of freedom are added to the nuclear part. Our framework enables accurate geometric integration of both the nuclear and electronic degrees of freedom yielding a time reversible and energy conserving dynamics on the ground state Born-Oppenheimer potential energy surface that is stable also under approximate self-consistent field convergence. Extended Lagrangian Born-Oppenheimer molecular dynamics provides a surprisingly simple and general framework for atomistic simulations. [A. M. N. Niklasson, Phys. Rev. Lett. vol 100, 123004 (2008); P. Steneteg et al., Phys. Rev. B vol 82, 075110 (2010)]

We present a framework, called Multiscale Analysis Platform (MAP), that serves to classify multiscale simulation methods into a set of motifs with distinct features. The classification scheme is based on factors such as the degree of coupling between scales, the extent of spatial and temporal scale separation, self similarity, and the relaxation times at the finer scale. We will introduce the MAP and discuss the framework in the context of problems relevant to materials scientists, such as materials for energy storage and catalysis.

Theories of protein crystallization based on spheres that form close-packed crystals conclude that assembly is enhanced by the metastable liquid-vapor critical point and is best within a defined `slot' of second virial coefficients. However, most protein crystals are open structures stabilized by anisotropic interactions. Here we use theory and simulation to show that the assembly of one such structure is not enhanced by the critical point or predicted by the second virial coefficient. Instead, optimal assembly requires the avoidance of liquid-vapor phase separation, and the imposition of a thermodynamic driving force on the order of the thermal energy.

The real-space implementation of the Kohn-Sham density-functional theory (RSDFT) using the finite difference method for the derivative of variables, has attractive features of parallelizability in addition to universality in target materials and conditions. Following the divide-and-conquer (DC) strategy, we propose a linear scaling RSDFT method for various materials by advancing the former formulation in Ref. [1]. In the Kohn-Sham-type equation for a domain, we introduce (1) the density-template potential for the density continuity at the domain boundary with simple stepwise weight-functions and (2) the embed potential to take into account all the quantum correlation effects with other overlapping domains in addition to the classical effects of ionic and electronic Coulomb potentials [2]. We thereby realize reasonably high accuracies in atomic forces irrespective of the electronic characters of materials. The timing tests on parallel machines demonstrate the linear scaling of the the DC-RSDFT code with little communication time between the domains. The DC-RSDFT method mentioned above is hybridized with the classical, empirical inter-atomic potential model using the buffered cluster method [3]. In the hybrid quantum-classical method, we apply the DC-RSDFT method to a region in which chemical reactions should occur, while the classical empirical atomistic method to the environmental region. The mechanical coupling of the two regions at the atomistic scale is realized by introducing artificial buffer atoms to both methods. The atomic forces are calculated concurrently at every time step to simulate the dynamics of the atoms. We apply the hybrid quantum-classical simulation method to investigate the Li diffusion processes in both graphite [4] and graphite-electrolyte interface. References [1] F. Shimojo, R. K. Kalia, A. Nakano, and P. Vashishta, Comp. Phys. Commu. 167, 151 (2005). [2] N. Ohba, S. Ogata, T. Kouno, and T. Tamura, Comp. Phys. Commu., submitted. [3] S. Ogata, Phys. Rev. B 72, 045348 (2005). [4] N. Ohba, S. Ogata, T. Tamura, S. Yamakawa, and R. Asahi, Comp. Model. in Eng.&Sci. 75, 247 (2011).

Understanding the adsorption of complex molecules on solid surfaces is an essential element of materials design. Both energetic and entropic factors may play a role, and molecular simulation is a valuable tool to investigate the free energy landscapes that influence the thermodynamics and kinetics of adsorption. While detailed first-principles modelling, and atomic-scale simulations using accurate interaction potentials, are vital, it is also necessary to turn to more coarse-grained models, in order to fully explore the available configuration space and obtain general physical insights. Enhanced sampling techniques, such as the Wang-Landau method and parallel tempering, enable us to obtain a very detailed statistical mechanical picture, identifying phase transitions and ground states. This talk will briefly review recent work in the area, and describe some of our results on coarse-grained models of polymers and peptides in confined geometry.

Layered mineral composites have a substantial potential impact in areas such as energy applications (oil industry additives), materials applications (nano composites materials) and biomedical applications (drug delivery). The microscopic structure and mechanisms of layered nanomaterials operate over many different length scales, ranging from nanometers to microns; one of the key challenges in the simulation of such systems is efficiently sampling these scales to understand how the microscopic structure affects the macroscopic properties of the composite. For example, the mechanical enhancement of clay-polymer nanocomposites depends on factors over various length scales: the orientation of the whole clay platelet in the polymer matrix will affect the mechanical resistance of the composite, while at the shortest scale the molecular arrangement and the adhesion energy of the polymer molecules in the galleries and the vicinity of the clay-polymer interface also affect the overall mechanical properties. In this talk, we address the challenge of creating a hierarchal multiscale modelling scheme to traverse a sufficiently wide range of time and length scales. This scheme uses separate simulations at one length scale in order to pass the acquired input parameters to higher length scales, starting from the electronic structure through classical atomistic molecular dynamics to coarse-grained models. For example, the interactions between the organic and inorganic phases are derived from more accurate levels: the partial charges of the atomistic atoms are derived from a fitting procedure for periodic systems to the electrostatic potential and the coarse-grained interaction potentials are calculated by matching structural properties with the atomistic simulations. We also provide reverse-mapping from the coarse-grained models onto the atomistic structures and from atomistic to electronic structure calculations which provides a double function; the higher levels of the scheme can be used to increase the sampling of the more accurate methods, and refinement of the interaction potentials for coarser levels can occur "on-the-fly". Such a scenario is well suited to distributed computing with each level of the scheme allocated to a suitable computational resource. We shall describe how tools developed by the MAPPER project (Multiscale Applications on European e-Infrastructures, www.mapper-project.eu) facilitates our multiscale scheme. Using this new technology, combined with unprecedented resources through geographically diverse computational grids, we have simulated several clay-polymer systems containing up to several million atoms/particles. This system size is firmly within the mesoscopic regime, containing several clay platelets with the edges of the platelets explicitly resolved. We show preliminary results of a â?obottom-upâ? multiscale simulation of a clay platelet dispersion in polyethylene glycol and the effect on mechanical resistance.

For the computational study and design of devices incorporating multi-domain ferroelectric materials, it is necessary to extend the current capabilities of the phase field model up to the micron scale where experiments are typically performed. Also arbitrary electrical and mechanical boundary conditions need to be incorporated relatively easily. In this work, we report a time domain implementation of the 3D phase field model that can simulate multi-domain ferroelectric switching in devices including heterogeneous materials. This massively parallel implementation enables us to study the switching properties of micron size devices with ~10⁹ degrees of freedom. We used a mixed finite difference and finite element grid, for calculating the nonlocal electrostatic and elastic interactions respectively. Special assembly and solver algorithms were implemented in order to avoid non-local assembly and computation in distributed computing architecture. All the local and non-local interactions are shown to scale linearly up to thousands of processors. The model can take into account arbitrary electrical and mechanical boundary conditions suitable for the study of devices with arbitrary structures. Using this model, we report the simulation results of ferroelectric switching in devices incorporating the multi-ferroic material BiFeO₃. We explain the nucleation and domain growth mechanism observed in multiple experiments reported recently on various surfaces of BiFeO₃. Particularly, we show the importance of electrical boundary condition and correlated nucleation mechanism that determines the domain evolution under an external electric field. Finally, we show the result of domain switching under transverse applied field that led to the demonstration of purely electric field induced magnetization reversal in the BFO/FM heterostructure.

A nondeterministic multiple scale approach based on numerical solution of the Monte-Carlo master equation on atomic lattices solved together with a standard finite-element formulation of solid mechanics is discussed. The approach is illustrated in application to long-term evolutionary processes of volume diffusion, precipitation and creep cavity self-healing in nanocrystalline austenite (Fe fcc) samples. A two-way mechanokinetic coupling is achieved through the implementation of strain-dependent diffusion rates and dynamic update of the finite element model based on the atomic structure evolution. Strain dependency of the transition-state theory diffusion rates is discussed as a key element of the present methodology. The effect of macroscopic static loading and cavity geometry on the total healing time is investigated for the material of interest and found to be in compliance with the experimental observations. Meanwhile, the modeling effort takes analysis far beyond the available empirical data and allows quantitative predictions of various long-term characteristics, such as total healing time, hardly accessible by experimentation and their dependence on damage profile and loading conditions. The approach is widely applicable to the modeling and characterization of advanced functional materials with evolutionary internal structure, surface properties and synergistic behavior in material systems.

Mechanisms involving crystal point (vacancies and impurities), line (dislocations), plane (grain boundaries, interfaces and cracks), and volume (precipitates, twins) defects define strength and deformability of crystalline materials. In particular, experimental and theoretical investigations carried out in the past have ascertained the role of dislocations formation, motion, and interaction with other defects in defining both the plastic yield and the hardening behaviour in crystalline structures. The discrete nature of defect interactions suggests that characterisation of plasticity should involve a truly multiscale framework [1,2]; furthermore, there exists an intimate coupling between the disparate length scales, e.g. long-range stress fields can cause the nucleation of new defects or drive the motion of existing ones, which in turn affect the macroscopic plastic flow. Therefore, although hierarchical multiscale frameworks are often used, whereby length and time scales are not coupled and the communication between smaller and larger scales is unilateral (or one-way, e.g. going from smaller to larger scales), a concurrent framework should be invoked to achieve a bilateral communication between the physical models implemented at different scales [1]. In this paper, we focus on a description of plasticity ranging from the length and time scales of discrete dislocation dynamics (nm/µm and ns) to those of a crystal plasticity continuum mechanics formulation (µm/mm and s). In particular, we concentrate on the coupling between dislocation dynamics (DD) and crystal plasticity finite element (CPFE) models in order to treat micrometer length scales in realistic test configurations and loading conditions of relevance for a number of engineering applications. The proposed approach is based on submodelling and employs an iterative procedure aimed at satisfying displacement and traction compatibility at the submodel boundary. After testing and validating the algorithmic development in [3], concurrent coupling DD/CPFE simulations of a FCC polycrystalline microfilms subjected to tension and indentation are presented here. Although other concurrent multiscale frameworks have been very recently used for the coupling of dislocation dynamics and crystal plasticity formulations [4], the present model is the first attempt to use submodelling and a CPFE description implemented in a commercial finite element package, which should greatly increase the usability of the model. References [1] Liu, W.K., Karpov, E.G., Zhang, S., and Park, H.S., Comput. Methods Appl. Mech. Eng., 193 (2004): 1529-1578. [2] Curtin, W.A., and Miller, R.E., Modell. Simul. Mater. Sci. Eng., 11 (2003): R33-R68. [3] Balint, D.S., Dini, D., Zografos, A., and Zhu, T., Proc. EUROMECH Colloquia: Multiscale effects in fatigue metals, 5-9 July 2010, Paris, France. [4] Wallin, M, Curtin, W.A., Ristinmaa, M., and Needleman, A., J. Mech. Phys. Solids, 56 (2008): 3167-3180.

This talk will discuss theoretical and computational tools developed in our research group for the last several years. Of particular importance are screened hybrid functionals like HSE, HISS, and their successors based on local range separation. I will also present some of our current developments based on Projected Quasiparticle Theory.

One of the main goals of a materials scientist is to be able to design materials with desired properties, helping to advance our technological society. Better medical drugs for treating diseases without side effects, better batteries for longer lasting computers and wider range electrical cars, safer nuclear reactors with less radioactive waste, and higher temperature superconductors for more efficient power distribution, are but a few examples. In heterogeneous systems, intuition is often not a reliable guide, and the difficulty in synthesizing new complex materials limits the usefulness of the experimental approach. In order to search for new materials with desired functionality, an accurate computational capability is thus highly desirable. Because no experimental results exist, this capability must have reliable predictive power. At a minimum, trends in properties with respect to material composition need to be accurately captured by any such computational approach. While extreme-scale computation has given us hope that we will someday be able to accurately model the structural composition of complex heterogeneous materials and then calculate their properties, computing power is not enough. Without a truly predictive set of equations supporting this capability, the computational approach will fail. My work is focused on improving the predictive capability of the underlying equations. There are very accurate and predictive capabilities in use in Quantum Chemistry calculations, such as the Coupled Cluster expansion. Even with extremely large super computers, however, the computational cost of these methodologies is prohibitive. This is an illustration of the constant competition between accuracy and computational cost in this field. Density Functional Theory (DFT) and its time dependent implementation (TDDFT) have the formal theoretical requirements for supporting computationally efficient computational approaches. However, presently available approximations for the exchange-correlation functional, limit the predictive power of this approach. In this talk, I will discuss my approach to designing new functionals, and my work on creating functionals for van der Waals' bonded systems, such as explosives, and materials containing elements with d- and f-electrons, with transition metal oxides as an example. Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000.

Halogen terminated surfaces are more reactive alternatives to the hydrogenated silicon surface for functionalization reactions. However, despite the widespread use of chlorinated silicon as the starting point for further functionalization reactions, the high reactivity of this surface towards simple polar molecules such as NH3, H2O, H2S still remains unclear. Therefore, we undertook a comprehensive density functional theory investigations to elucidate the factors that govern the reactivity of fully and partially halogenated Si(100) and Si(111) surfaces. We considered Cl and Br. Both electronic and steric factors explanin the observed reactivity trends. We show that steric effects (more important for increasing halogen coverage or for increasing halogen size) may be used to tailor the surface reactivity. The halogenated surfaces show a considerable activation with respect to the hydrogenated surface. The reaction on the halogenated surfaces proceeds via the formation of a stable dative bonded complex in which a silicon atom is pentacoordinated [1]. The activation of the halogenated Si(100)-2Ã-1 surfaces towards the polar molecules arises from the large redistribution of charge in the transition state which precedes the breakage of the Si-X bond -. Steric effects also play an important role on the surface reactivity, making brominated surfaces less reactive than the chlorinated surfaces. The voluminous bromine atoms prevent the nucleophilic attack of ammonia to the silicon atom of a SiBr group, which leads to an increase in the activation energy barrier. Therefore, the surface reactivity can be tailored by either changing the nature of the halogen atom or by changing its surface coverage. These results explain the fact that activation energy barriers are higher on the chlorinated Si(111) surface than on the chlorinated Si(100) surface because the 111 surface has a higher density of chlorine atoms than the 100 surface. [1] F. A. Soria, E. M. Patrito, P. Paredes-Olivera. Langmuir, 2011, 27, 2613.

Most currently used solar-absorbing photovoltaic (PV) materials such as Si, GaAs, and CuInSe2 have been discovered accidentally, and were subsequently improved incrementally over tens of years, at significant R&D cost. Indeed, many materials among those that have already been reported in databases (e.g., ICSD) may qualify as good PV absorber. To screen out those candidate PV materials, the current selection principles such as Shockley-Queisser limit efficiency depending only on material's band gap (no matter whether it is direct or indirect), have proven over the years to be grossly insufficient. Here we developed a criterion which incorporates material specific absorption as well as material specific radiative recombination loss. These quantities were evaluated by first principles quasi-particle calculation. The high-throughput calculations of ~500 I-V-VI materials and ~300 I-III-VI materials have identified over 40 potential PV absorber materials. The emerging design principle on how to improve the absorption strength near the threshold will also be discussed from data-mining.

Metallic nanoparticles and their arrays offer fascinating optical properties, originating from plasmon resonance phenomena. The experimental control over the size, shape and relative arrangement of such nanoparticles, along with the strong dependence of their optical properties on these structural parameters, offer potentially a range of interesting applications. Given, however, that the parameters space is vast, realizing this potential requires a method other than a trial and error approach for choosing the optimal set of structural parameters. We develop mathematical tools to design plasmonic nanoconstructs with predetermined optical properties and hence desired functionalities. Our approach is based on the combination of optimization algorithms with a finite-difference time-domain solution of the Maxwell equations, and is thus entirely general. We illustrate the applicability of the approach on the one hand to develop new insights into plasmonic physics, material properties and the interaction of near fields with matter, and on the other hand to design imaginative constructs with promising properties. In the talk I will first introduce the properties and potential of nanoplasmonics, placing this research in context with world-wide efforts on related themes, and next very briefly describe our numerical method before proceeding to discuss our results. I will conclude by extending the discussion to the challenges of guiding information in the nanoscale and of controlling the translational modes of molecules, via optimally-designed metal nanoparticle arrays.

Significant advances in Materials Science have been achieved by harnessing specific functionalities of nanostructures, such as improved mechanical, electrical and thermal properties, for particular applications. Predictive ab initio calculations suggest that designer nanostructures, such as schwarzites[1] and related foam structures[2] of carbon, may combine low gravimetric density with high stiffness and favorable electrical as well as thermal conductivity. Unusual charge and thermal transport properties can be expected in peapods consisting of doped fullerenes or diamondoids enclosed in a carbon nanotube. Successful synthesis of such nanostructures precludes detailed understanding of their microscopic formation mechanism. Combination of molecular dynamics simulations and total energy calculations provide guidelines to achieving chirality selective synthesis of carbon nanotubes without metal catalyst or the formation of unusual nanostructures on carbon saturated metals. Since direct observation of such atomic-scale processes is very hard by experimental means, computer simulations are a welcome alternative to gain microscopic insight into the underlying processes.
[1] S. Park, K. Kittimanapun, J.-S. Ahn, Y.-K. Kwon and D. Tomanek, J. Phys.: Condens. Matter 22, 334220 (2010).
[2] K. Umemoto, S. Saito, S. Berber, D. Tomanek, Phys. Rev. B 64, 193409 (2001).

Multilevel modeling play a key role in studying materials design for energy storage devices. A model which connects atomistic and mesoscales was proposed to study battery performance from first principles. Coupled ion and electron transport in nanostructured energy storage materials was modeled using collective long-range interactions and short-range effects of the finer-scale models. Classical density functional theory coupled with the Poisson_Nernst_Planck formalism were used to link the atomistic and mesoscopic length scales. Also it was proposed that three-dimensional description is essential for accurately reproducing experimentally measured conductivity in materials with complex nonlinear charge transport pathways. Consequently, the computational efforts required may increase dramatically and scalable parallel computation designs are needed. In this study, we present a parallel processing approach based on PNP cDFT method for hierarchical multiscale modeling. There are two major computations involved in the hierarchical multiscale modeling, i.e. evaluation of the chemical potentials of charged species using the classical density functional theory, and solving Poisson_Nernst_Planck equations by Newton-Raphson method. To efficiently compute chemical potentials, the computational steps are separated into different modules. Redundant computation is removed through modulization and a parallel version of Picard method of successive approximation is used. For solving the equations, modulization steps are also used to remove redundant computation and a parallel Jacobi method is used. Analysis of the problem shows that only boundary values need to be communicated between different loops, which make Jacobi implementation suitable for solving the problem. Possible speeding up strategies of parallel Jacobi method, e.g. the successive over relaxation scheme (SOR), is also investigated. Numerical results on coupled ion and electron transport in nanostructured energy storage materials using the developed parallel PNP cDFT method demonstrate the multiscale modeling capability of the developed method. The scaling of the method with respect to processors in high performance computing environment is also studied.

The intricate coupling between the different levels of organisation in multiscale systems is very common in diverse fields such as material science, biochemistry, and medicine but also extremely hard to uncover, particularly as a result of the need to simplify the level of description. Recently, a powerful theory called Stability [1,2] has been shown to have the capacity to unravel all the relevant scales in such systems without the need to coarse-grain their description, thereby entirely preserving the coupling between scales. Using a natural network representation of the system at its finest scale, it identifies a hierarchy of partitions of the system from the behaviour of a dynamical process taking place on the graph. The high requirements in computational resources for treating multiscale systems generally impose strong limitations. This computationally efficient approach overcomes the usual trade-off between detailed description and access to high scales by probing the dynamics at all scales simultaneously, while at the same time preserving all the details at the finest level of description. The presence of multiple scales is notably an inherent property of protein assemblies, reflected in both their many levels of structural organisation as well as in their wide variety of motions, spanning around 15 orders of magnitude in time. Consequently, Stability has been applied on various types of proteins, including common enzymes such as adenylate kinase, large multimers such as rubisco, recently discovered structures such as the myosin tail-interacting protein from the malaria parasite. From the comparison with experimental data from X-ray crystallography, NMR spectroscopy, and binding assays, the hierarchy of the identified partitions has been shown [3] to encompass the different types of motions and structural levels of organisation, and provide detailed information on the mechanisms underlying the biological function, and how they emerge from the biochemistry at the smallest scales. [1] Delvenne, J. C., Yaliraki, S. N., and Barahona, M. (2010) Proc. Natl Acad. Sci. USA 107(29), 12755â?"60; arXiv:0812.1811v4 [physics.soc-ph] (2008). [2] Lambiotte, R., Delvenne, J. C., and Barahona, M. Laplacian dynamics and multiscale modular structure in networks arXiv:0812.1770v3 [physics.soc-ph] (2009). [3] Delmotte, A., Tate, E. W., Yaliraki, S. N., and Barahona, M. (2011) Phys. Biol. 8(5), 055010.