The Multi-paradigm Multi-scale Simulation Facility CMDF: An Introduction and Overview

Transcription

1 The Multi-paradigm Multi-scale Simulation Facility CMDF: An Introduction and Overview Markus J. Buehler Materials and Process Simulation Center (MSC) Division of Chemistry and Chemical Engineering California Institute of Technology 1 st Workshop on Multi-Paradigm Multi-Scale Modeling in the Computational Materials Design Facility (CMDF), Aug. 23/24, 2005

2 Brittle versus ductile: How atomistic interactions govern materials behavior glass ceramics metals copper (Rice, Thompson, Peierls, Beltz, Yip, Argon et al.) Different materials respond very differently to application of mechanical loading. Goal: Understand relationship of the overall mechanical response and the interatomic interaction, ultimately governed by quantum mechanical concepts

3 Example: Brittle fracture energy flow engineering scale ohesive laws quantum mechanics Supersonic fracture (Buehler et al.) In brittle fracture, the macroscopic behavior of the materials depends on its underlying atomic interaction across several hierarchies of scales

4 Mechanical behavior of different kinds of materials brittle : Materials that experience little, if any, plastic deformation before the onset of fracture ductile : Materials that experience significant plastic deformation before the onset of fracture (Buehler et al., Nature, 2003) geometric confinement Nanostructured materials, carbon nanotubes Use large-scale computing within framework of multiscale modeling Develop fundamental understanding (Buehler et al., CMAME, 2004) biological materials (Proteins, DNA ) (Buehler et al., JMPS, 2002) + other properties (optical, electronic...) (Buehler et al., MRS Proceedings, 2004)

5 The scale coupling problem Want: Accuracy of quantum mechanics (QM) in atom systems This is impossible (today and in the foreseeable future) Possible solution: Multi-scale modeling techniques based on hierarchies of overlapping scales ~10 23 atoms atoms CMDF Concept: finer scales train coarser scales

6 The vision for multi-scale modeling Long-standing dream Calculate macroscopic properties of materials by theoretical modeling or computer simulation from a very fundamental, ab initio perspective The solution is in reach: Achieving this goal may be possible within a timeframe of the current decade. This progress is possible with The advent of efficient and accurate quantum mechanical methods (e.g. DFT), Development of new empirical and semi-empirical potentials (EAM, ReaxFF ), Enormous growth of computing power enabling studies with billions of particles. Critical: Breakthroughs in scale coupling techniques (e.g. QC method) and analysis methods for complex systems (centrosymmetry technique) Vision: Atomistic simulations of engineering properties at macroscopic scales to 1) understand fundamental mechanisms in materials (e.g. deformation, assembly), and to 2) predict properties of new materials to design novel materials (DARPA PROM)

7 Impact of multi-scale modeling Multi-scale modeling may have major impact in particular in the engineering disciplines Realization: To understand how materials behave, perform, fail, assemble it is critical to appreciate the atomistic viewpoint (optimize materials for utilization at their fullest extent..) In classical application of macroscopic materials (e.g. steel, alloys): Understanding, knowledge of how material properties derive from atomic scale is critical to improve the performance of existing materials New technologies: In particular, in newly developing nano-technologies (N/MEMS), bioinspired materials and technologies (e.g. artificial silk based materials, new smart hydrogels), metallic glasses: Knowledge about the atomic structure-function/response relation is critical to engineer and design the materials (can not even apply continuum theories immediately due to e.g. small volume)

8 Outline Motivation (previous slides): From small to large Introduction: Multi-scale multi-paradigm concepts Overview over CMDF Examples of CMDF multi-scale simulations Discussion and conclusion Overview: Workshop program (remarks)

9 (Molinero et al.) Concurrent versus hierarchical glucose monomer unit multi-scale simulations DNA CMDF (Buehler et al.) Glycosidic bond atomistic and M3B meso model of oligomer (Pascal et al.) QC Hierarchical coupling finer scales train coarser scales Concurrent coupling (QC-Tadmor, Ortiz, Phillips, MAAD-Abraham et al., Wagner et al.) Spatial variation of resolution and accuracy

10 What are multi-paradigm simulations? For accurate prediction of materials properties based on first principles, need seamless integration of codes and methods ranging from quantum mechanics up to the continuum scale Most simulation methods: Only allow usage of a single method (e.g. EAM, DREIDING FF, QM ), rather than spanning a whole spectrum of state-of-the-art tools, including visualization & analysis A multi-method environment that allows scale-/ paradigm-agnostic combination of various simulation engines and simple additions to codes (e.g. new BCs) does not exist We propose: Development of multi-method multi-scale simulation tools allowing for seamless combination of various methods Computational Materials Design Facility (CMDF)

11 Single-paradigm versus multi-paradigm modeling In In E D CODE (numerical method) Out A C B F Out Single paradigm Multi-paradigm (becomes multi-scale when A,B operate at different scales) paradigm complexity 1) Computer science component (code coupling), interface 2) Strategic component (multiple simulation paradigms to achieve goal)

12 Main features of CMDF CMDF allows Multi-scale modeling Multi-paradigm modeling Easy to install Easy to use Scale agnostic (works at various scales) Extensible: Easy to modify and to advance (easy implementation of new methods) Goal: Allow to focus on science rather than code development

13 The principal objectives of CMDF To provide a very general, extensible approach of a simulation environment: Have library of a variety of computational tools spanning the scales To establish a re-usable library of advanced computational tools that can be used as black boxes for most applications, and that initialize themselves with standard parameters for easy usage in standard cases. To enable atomistic applications to be used by engineers and experimental scientists, while retaining the possibility of building highly complex simulations and models, educational component To close the gap in coupling fundamental, quantum mechanical concepts such as DFT to empirical force field descriptions such as DREIDING, UFF and other nonreactive force fields by using the ReaxFF force field. To provide a test bed for new model and algorithm developments, and making it simple to develop new communication channels between computational engines (e.g. in development of new force fields composed out of distinct methods such as QEq Morse potentials, ReaxFF or many others). To represent an intellectual home and a forum for discussion for computational methods, techniques, as well as for novel philosophies of scale coupling and their implementation.

14 The CMDF integration strategy The development of CMDF bases on the concept of calling simulation modules (libraries) from a Python scripting environment This technology allows reuse of existing codes within a modern scripting environment without touching the individual source codes Python modules Method 1 QM Method 2 ReaxFF CMDF framework Method 3 DREIDING Method 5 FEM Method 4 Mesoscale

15 CMDF fact sheet Ca. 35 different modules for various tasks including File and data I/O (readers, file writers) Energy and force calculation at different levels of detail Pre- and post-processors (builders, analysis methods) Filters and glues between different methods and scales GUI for visual building... and much more CVS based code development and checkout Effort started around 2000 by Rick Muller then carried on by Peter Meulbroek until 2004 and taken over by Markus Buehler in 2004 Underlying codes in Python, C/C++ and FORTRAN codes seamlessly combined through wrappers, scripts and object code level integration often usingh Python Various modules can be used to build complex simulation tasks, such as multiple force field concurrent simulations, multiple scale/paradigm simulations, Still under development but during the last year we have reached critical number of working (and interfacable ) modules so that CMDF can be applied to real scientific problems

16 CMDF development strategies All developments are driven by specific scientific problems to be solved Academic code development is only succesful with a clear scientific mission and application The usefulness of CMDF will be determined by the community s need for the new methods: This can be encouraged or demonstrated by solving outstanding science problems Other multi-scale integrated frameworks: CampoASE, MMTK, LAMMPS, Cerius2... CMDF is a most general approach with a clear scale-agnostic mission

17 The Caltech CMDF Team Markus (CMDF team leader, director of multi-scale modeling at MSC) Jef (staff; Python coding, executable wrapping, SWIG, global data model) Hatem (RA; quantum mechanics, LE4 etching, Jaguar wrapping, dynamics) Barry (staff; force fields modules Molscape, QEq, FORTRAN coding) Frank (graduate student; solvation methods, APBS) Tod (graduate student; solvation methods, APBS) John (graduate student; bio-methods, Docking procedure) Qingsong (post doc; pqeq force field for ferroelectrics, C++/C) Julius (graduate student; Gaussian charge module) Li Tao (undergraduate student; coupling DREIDING-ReaxFF for enzymatic reactions) Sergey (post doc; reactive force fields REBO) Victor (graduate student; SCREAM) Daniel Yi (undergraduate student; time scale extension, TAD methods) Yves Lansanc (ModSim) Frank (APBS continuum solvation) Chris George (I-V calculations) Si-ping (graduate student; ReaxFF development and modularization, Fortran) Adri van Duin (director; ReaxFF and force field technology at MSC) Vaidehi Nagarajan (director, biotechnology, MSC)

18 The CMDF Team outside of Caltech The JPL team includes Paul von Allmen (group leader, high performance computing group at JPL, EZTB tight binding codes) Fabiano Oyafuso (post doc; calculation of thermoelectrical properties using CMDF) Lei Pan (staff; parallelization of CMDF) Joey Czikmantory (staff; WIGLAF internet hookup for CMDF) Seungwon Lee (staff, thermoelectrical properties of Bi2Te3, Sb2Te3) Alberto Cuitino (Rutgers Univ.) + group members THANKS TO ALL CMDF DEVELOPERS!

19 Core strength of CMDF: Integration of reactive potentials into multi-scale scheme DFT: ~500 atoms (ps) ReaxFF: ~10,000 atoms (ns) (needed for realistic description of materials/chemistry) There exists a huge gap in performance between QM and empirical MD ReaxFF (van Duin et al.) is a possible bridge (fully first principles trained)

20 Motivation: Need for reactive potentials The scope of materials and properties accessible to atomistic modeling is currently limited to relatively simple atomic microstructures (chemistry plays minor role). Reason: Although numerous empirical interatomic potentials exist that can describe thermodynamic equilibrium states of atoms, so far, all attempts have failed to accurately describe the transition energies during chemical reactions using more empirical descriptions than relying on purely quantum mechanical (QM) methods. However: Huge scale gap between QM methods and empirical descriptions of materials q q q A?? A--B Large-strain properties close to bond breaking... q q q q q q A--B B A B Attempt: Using Morse potential for bond breaking (can be dangerous) van Duin et al.

21 ReaxFF: A reactive force field in CMDF E system = E bond + E vdwaals + E Coulomb + E val + E tors sp3 + E over + 2-body E under multi-body 3-body 4-body sp2 sp A bond length/bond order relationship is used to obtain smooth transition from non-bonded to single, double, and triple bonded systems. All connectivity-dependent interactions (i.e. valence and torsion angles) are made bond-order dependent Ensures that their energy contributions disappear upon bond dissociation Feature non-bonded interactions (van der Waals, Coulomb): Shielded ReaxFF uses a geometry-dependent charge calculation scheme (similar to QEq) that accounts for polarization effects Breakthrough: bridge between QM and empirical FFs

22 Summary: Various simulation methods in CMDF

23 Without central data structure Status BEFORE No central data structure: Needed individual translators to communicate between different methods/modules No generic, flexible approach to method coupling Method 1 QM Method 2 ReaxFF Translater between method 1 and 3 Translater between method 2 and 5 Translater between method 4 and 1 Translater between method 2 and 4 Method 3 DREIDING Method 5 FEM Python modules Method 4 Mesoscale Jef Dodson et al.

24 Development of the central data structure based on Extended OpenBabel (XOB) Status NOW Have central data structure based on X OpenBabel (XOB) Allows generic, flexible and extensible approach to method coupling Method 1 QM Method 2 ReaxFF Central data structure Extended OpenBabel XOB Method 3 DREIDING Method 5 FEM Python modules Method 4 Mesoscale Talk by Jef Dodson on the XOB data structure

25 CMDF: A typical script OBMol=tools.readbgf ( water.bgf ) CMDF module BGF file reader CMDF module Builder tools tools.insertcrack(obmol, 10, 20, 2.3) CMDF module CMDF module CMDF module Preprocessing (e.g. Typing) Dynamics Integration Analysis Data processing tools.forces(obmol, EAM, ReaxFF,...) dynamics.integratenve(obmol,...) OBMol=tools.build_MWNT (10, 8, 4) OBMol=tools.readbgf ( water.bgf ) OBMol=tools.build_MWNT (10, 8, 4) tools.insertcrack(obmol, 10, 20, 2.3) Loop tools.forces(obmol, EAM, ReaxFF,...) dynamics.integratenve(obmol,...) tools.printfracture (OBMol) Python scripting OBMol: ModBabel global data structure

26 Example: Simple replacement of different methods OBMol=tools.readbgf ( water.bgf ) OBMol=tools.build_MWNT (10, 8, 4) tools.insertcrack(obmol, 10, 20, 2.3) tools.forces(obmol, EAM) or tools.forces(obmol, ReaxFF) Loop tools.forces(obmol,???) dynamics.integratenve(obmol) or tools.forces(obmol, Tersoff) tools.printfracture (OBMol) or tools.forces(obmol, JAGUAR DFT X,B3LYP) Our scale agnostic design using the central data structure allows simple replacement of different simulation methods: Design methods that operate on OpenBabel objects This can be used to try different simulation engines during a simulation to find the most efficient method required to achieve a desired accuracy (concept of virtual dynamics; will be developed in phase 2)

27 Brief summary of applications of CMDF in describing complex materials Oxidation of metals and competition with crack extension Chemo-mechanical coupling Cracking of silicon Technologically relevant for N/MEMS Computational strategies to extend accessible time scales TAD method integrated in CMDF Highlights of the CMDF workshop Massively parallelized computing (JPL) ReaxFF/MM scheme for modeling enzyme activity (Caltech) Sensitivity analysis of macro-scale properties w.r.t. to accuracy of QM results (JPL) WIGLAF Internet hookup for CMDF (JPL) TWiki site for CMDF

28 Example for code coupling: Concept of mixed Hamiltonian Method A (encompasses defect) Method B Concurrent multi-scale modeling to focus on localization of defects, failure...

29 Example for potential coupling: Concept of mixed Hamiltonian ( handshake ) Developed scheme to couple different codes with each other based on weights describing the amount of force and energy contribution of different force engines: Works well for certain force fields (linear and sinusoidal) Method A Method B

30 Reactive versus non-reactive potential nonreactive Energy Reactive nonreactive reactive Bond separation

31 Example applications Succesfully applied to... Metal/organic systems and cracking (EAM-ReaxFF) Markus, Si-ping Proteins (DREIDING-ReaxFF) Li Tao Alloys (RGL-ReaxFF coupling) Peng Xu, Qing Oxidative processes (EAM-ReaxFF) Markus Organic molecules (ReaxFF-DFT) Hatem... And other systems Example: ReaxFF/MM scheme (Li Tao) all the details in individual talks

32 Oxidation of a metal (Al) surface Examples demonstrates the concept of the moving boundary between different computational engines Boundary location determined by position of oxygen atoms Iterative procedure to find Φ Φ

33 Deformation of Al and Al oxidation Mode I y x Nano-void in Aluminum filled with O 2 : Load applied in the x- direction The system is under 10% strain in the x-direction (orthogonal to the long axis of the elliptical defect). The results demonstrate our capability to couple complex chemistry competing with fracture events.

34 Deformation of Al and Al oxidation Close-up view of the interface between the EAM and the reactive region.

35 Deformation of Ni and Ni oxidation 10% 12.5% 15% 20%

36 Deformation of Ni and Ni oxidation 10% loading

37 Deformation of Ni and Ni oxidation 15 % strain y x

38 Al versus Ni 10 % strain (x-direction), T=400 K After ca. 10 ps Al Ni

39 Cracking in Silicon: Pure ReaxFF Mode I loading Drawback: System size limited to about 1,000..2,000 atoms But: Mechanical properties typically require much larger system sizes (10K-100K) Use concurrent multi-scale method

40 Simulation Geometry: Cracking in Silicon We consider a crack in a silicon crystal under mode I loading. We use periodic boundary conditions in the z direction corresponding to a plane strain case. The smallest system contains 13,000 atoms and the largest system over 110,000 atoms. In the largest system, L x 550 Å and L y 910 Å. The number of reactive atoms varies between 500 and 3,000. Calculation of forces and energies in the reactive region is the most expensive part

41 Cracking in Silicon: Model within CMDF To model cracking in Silicon more efficiently, we developed a multi-paradigm scheme that combines the Tersoff potential and ReaxFF The ReaxFF region is moving with the crack tip (region determined based on local atomic strain) New hybrid scheme within CMDF (110) crack surface, 10 % strain Reactive region is moving with crack tip ReaxFF Tersoff CMDF reproduces experimental results (e.g. Cramer, Wanner, Gumbsch, 2000)

42 Cracking in Silicon: Hybrid model versus Tersoff based model Crack propagation with a pure Tersoff potential (left) and the hybrid ReaxFF-Tersoff scheme (right) along the [110] direction (energy minimization scheme). The snapshots are both taken with the same loading applied and after the same number of minimization steps. The systems contain 28,000 atoms and L x 270 Å and L y 460 Å. Conclusion: Pure Tersoff can not describe correct crack dynamics

43 Cracking in Silicon: Model within CMDF (110) crack surface, 20 % strain (100) crack surface, 10 % strain

44 Cracking in silicon: Movie Crack speed approaches about 75 % of the Rayleigh-wave speed (similar to experimental results) Small reactive region at crack tip (not shown explicitly here) 10 % strain in mode I applied; temperature T=300 K Ca ,000 reactive atoms embedded in classical region of >100,000 atoms (computationally very efficient)

45 Oxidation versus brittle fracture Crack dynamics in silicon without (subplots (a) and (c)) and with oxygen molecules present (subplots (b) and (d)) Subplots (a) and (b) show the results for 5 percent applied strain, whereas subplots (c) and (d) show the results for 10 percent applied strain. The systems contain 13,000 atoms and Lx 160 Å and Ly 310 Å.

46 Fracture of Carbon Nanotubes (CNTs) Tersoff ReaxFF Change of strength of CNTs as a function of presence of defects Compare to Griffith theories Confinement effects on fracture behavior Developed hybrid model of ReaxFF and Tersoff to model fracture of CNTs All bond breaking described by first principles ReaxFF

47 Deformation of materials at slow strain rates ε. Defect ε. Long time dynamics at crack tip using TAD Boundary conditions elasticity provided by large background system

48 Bridging to longer time scales Coupling TAD with ReaxFF in CMDF Talk by Daniel Yi H on Pt(100) surface Bridge site FCC site ReaxFF interfaced with TAD through CMDF (Collaboration with Art Voter, LANL)

49 Application: Cracking at low strain rates Cheap pair potential Continuum (FE) TAD ReaxFF EAM Hybrid scheme FE-EAM-ReaxFF-TAD enables achieving strain rates as low as 1% strain per microsecond (pure MD: 1% strain on the order of picoseconds) But: Have full atomistic complexity at crack tip while overall dimensions are on the order of micrometers

50 The multi-paradigm computational scheme Hybrid concurrent multi-scale model Dynamics Select subset region crack tip hotspots Integrate Subset Integrate subset region into large system Illustrates paradigm complexity (Buehler et al.) Perform time accelerated dynamics (e.g. TAD) Simulate for specified time or until N number of events (bond breaking/formation) happen; fixed BCs at transition layer

51 CodeA (communicator 1) Concept Technical Approach Bind MPI library calls to Python (The group handles and the communicators are INTEGERs in Fortran.) Use regular Python interpreter Use MPMD (Multiple Program Multiple Data): mpirun -p4pg pgfile to start a mixture of Python and Fortran/C processes Python and Fortran/C processes create MPI groups and communicators together Large-scale Applications of CMDF Parallelization with Python MPI The MPI universe (MPI_COMM_WORLD) CodeB (communicator 2) Objectives Glue together existing CMDF Fortran/C MPI parallel programs Develop scalable large-scale CMDF applications that fully utilize large-scale clusters Facilitate communication among separate MPI programs via efficient message passing rather than slow disk data transfer Use Python to minimize/avoid code changes that require recompiling Results Lei Pan CodeA and CodeB work as before except that they live in their respective groups with separate communicators (minor change from using MPI_COMM_WORLD to using a communicator passed in) Python code can reduce results from both CodeA and CodeB; or help direct communication across processors of CodeA and processors CodeB by using MPI_COMM_WORLD (no need to use disk files to transfer data between CodeA and CodeB) Python code can glue in future CMDF modules to build large-scale applications, without needing to recompile the existing modules

52 Hybrid modeling of enzymatic reactions in proteins Concept Results: ReaxFF/MM scheme Use combination of DREIDING and ReaxFF to model enzymatic reactions (focus on substrate binding site): Extensible to other systems Transition region between models allows smooth coupling Non-reactive region (Biograf, DREIDING) Reactive site with ReaxFF Contains substrate Objectives Provide new computational scheme to model influence of mutations on reactive barriers in enzymes Hybrid scheme of ReaxFF and DREIDING to allow for computational efficiency (focus on reactive zone) Conclusions Successfully implemented a ReaxFF-MM method in the CMDF framework Allows for rapid screening of different reaction paths and the influence of amino acid mutations Reasonable approach that may substitute the common QM/MM schemes due to several advantages (no difficulties how to terminate QM regions, faster, easily parallelizable, ) Li Tao

53 Sensitivity of Thermoelectric Property Simulations w.r.t. QM accuracy Goal: Determine the sensitivity of thermoelectric property simulations to the accuracy of first principles electronic structure calculations. General Approach: Use first principles result from Caltech to fit tight binding parameters Compute thermoelectric properties using a parametrized tight binding approach Compute variation of the thermoelectric properties due to variations in the first principles results Paul von Allmen Status: Fitting tight-binding parameters for Si, Ge, C, III/V, Bi2Te3 to experimental and first principles target material parameters Compute thermoelectric figure of merit using the relaxation time approximation and experimental lattice thermal conductivity Compute variation of thermoelectric properties due to the variations of the material parameters Technical Approach: Tight binding method with arbitrary symmetry and number of nearest neighbors Relaxation time approximation or Boltzmann equation solution for the transport properties Thermoelectric figure of merit: Electrical Conductivity Thermopower ZT = S2 σt κ e +κ L Relative Variations σ κ e EcX EcL EcX min min S m e l m et τ κ L Electron Thermal Conductivity Lattice Thermal Conductivity ZT

54 Centralized documentation for ongoing projects Simple and easy to learn text formatting rules that allows anyone the ability to edit pages Accessible from any web browser MSC TWiki (Hatem Helal and Daniel Yi) To register use the following login information Username: MscFriend Password: gwiki

55 WIGLAF: CMDF Web Interface Logical WIGLAF Application Standalone Application GUI Input Shell XSL Transformation Native Input ITAP-MD GUI Input Python Python C++ XML Binding EZTB Web Applet MD Output Materials Properties Joey Czikmantory

56 Summary: Major Achievements With development of CMDF: We have developed a solid framework based on a data model that represents an extensible, scriptable environment, We have integrated a variety of codes at different scales (including ReaxFF, some of the JPL codes), We have developed strategies how the codes interact (representing self documented communication channels between software components), We have treated hybrid ReaxFF-EAM systems of oxidations of surfaces that comprise of over 50,000 particles which allows it to simulate realistic systems, and We have demonstrated the usage of MPI parallelization

57 Outlook: Future work Applications of our new methods Focus Focus on code/scale coupling strategies (e.g. seamless atomisticmesoscale coupling; atomistic-continuum coupling) Sensitivity analysis (error bars): Understand how errors translate thoughout different scales Collaborations with other teams to addres new problems and to integrate new methods into CMDF

58 What are we trying to achieve with this workshop? Report to the audience what we have accomplished in the area of multiscale modeling at MSC, in particular within the CMDF framework. Areas of focus include: Sensitivity analysis (critical for predicting properties of novel materials from first principles), Mechanical properties (cracking, reliability, wear, fatigue, corrosion), Biological systems and structure predictions of membrane proteins as well as general structure prediction algorithms for proteins, Chemical reactions and catalysis (including enzymes), Thermoelectrical properties (calculating the ZT figure of merit from first principles), and General methodology development of scale coupling. Provide a forum for discussing various aspects of multi-scale modeling, in particular for establishing relationships outside the Caltech-JPL-Rutgers- USC PROM team. Intend to form an interdisciplinary team of researchers Determine needs for future developments and learn about problems we can address with our existing methods. Receive critical feedback to our work from outside Reach out to the community and offer our codes, methods and ideas developed within the CMDF framework, to develop CMDF into a generally used and accepted simulation tool.

59 Four sessions Session 1: Introduction We start with an overview over the Materials and Process Simulation Center (MSC) at Caltech. There will also be a brief introduction by Emily Abbott, the Associate Director of Corporate Relations (Caltech). We continue with an overview and outline of workshop agenda, including a historical perspective on Python in scientific computing. Session 2: Computational aspects: Data structures, performance and largescale computing Introduce the computational concepts of the CMDF framework, featuring wrapping methods, data structures, graphical interfaces (GUIs), WIGLAF internet hookups for CMDF, and parallelization for large-scale applications. Session 3: Scale coupling and applications: From QM to macroscale Describe techniques we use to achieve the coupling of scales. We start with coupling of QM to ReaxFF, new electron force fields eff, proceed to coupling of ReaxFF to empirical nonreactive force fields, and finally move on to mesoscale simulations and coupling to the continuum scale. Highlights of scientific examples include crack dynamics in silicon (relevant e.g. for N/MEMS technology), new first principles reactive force fields ReaxFF, properties of DNA nanostructures, and properties of ferroelectric materials. Session 4: Novel computational approaches and prediction of properties of complex materials In this session we present numerical computational approaches based on paradigm complexity and thus heavily utilizing the CMDF infrastructure and its modules. These examples serve the purpose to demonstrate that exciting science is enabled with our new methods based on the concept of creating paradigm complexity. We further provide several examples of how simple integration of various modeling paradigms can be achieved within CMDF, indicating the ease of use of CMDF for new codes and methods for novel applications.

60 Workshop outline: Overview (day 1) Introduction Computer science behind CMDF Scale coupling techniques in CMDF

61 Workshop outline: Overview (day 1) contd. Paradigm complexity: Use multitude of numerical tools Panel discussion

62 Workshop outline: Overview (day 2) Hands-on examples; scripts CMDF installations Unlike the first day, the second day is very interactive and without a fixed schedule We present whatever the audience requests

63 Some references* Knap, J. and M. Ortiz, An analysis of the quasicontinuum method. J. Mech. Phys. Sol., (9): p Knap, J. and M. Ortiz, An analysis of the quasicontinuum method. Journal Of The Mechanics And Physics Of Solids, (9): p Li, X.T. and E. Weinan, Multiscale modeling of the dynamics of solids at finite temperature. Journal Of The Mechanics And Physics Of Solids, (7): p Becke, A.D., Density-function thermochemistry. 3. The role of exact exchange. J. Chem. Phys., (7): p Springborg, M., Density-functional methods in chemistry and materials science. 1997: Wiley research series in Theoretical Chemistry. Rappe, A.K., et al., Uff, A Full Periodic-Table Force-Field For Molecular Mechanics And Molecular-Dynamics Simulations. Journal Of The American Chemical Society, (25): p Ercolessi, F. and J.B. Adams, Interatomic potentials from 1st principle-calculations - the force matching method. Europhys. Lett., (8): p Zimmermann, J.A., Continuum and atomistic modeling of dislocation nucleation at crystal surface ledges. 1999, Stanford University. Baskes, M.I., Embedded-atom method: Derivation and application to impurities, surfaces and other defects in metals. Phys. Rev. B, (12): p Baskes, M.I., Determination of modified embedded atom method parameters for nickel. Materials Chemistry and Physics, (2): p Bazant, M.Z., E. Kaxiras, and J.F. Justo, Environment-Dependent Interatomic Potential for bulk silicon. Physical Review B-Condensed Matter, : p Zimmerman, J.A., H. Gao, and F.F. Abraham, Generalized Stacking Fault Energies for Embedded Atom FCC Metals. Modelling Simul. Mater. Sci. Eng., : p Allen, M.P. and D.J. Tildesley, Computer Simulation of Liquids. 1989: Oxford University Press. Tersoff, J., Empirical interatomic potentials for carbon, with applications to amorphous carbon. Phys. Rev. Lett., (25): p Sefcik, J., et al., Dynamic Charge Equilibration-morse stretch force field: Application to energetics of pure silica zeolites. J. Comput. Chemistry, (16): p Mayo, S.L., B.D. Olafson, and W.A. Goddard, Dreiding - A Generic Force-Field For Molecular Simulations. Journal Of Physical Chemistry, (26): p Abraham, F.F., et al., Simulating materials failure by using up to one billion atoms and the world's fastest computer: Work-hardening. P. Natl. Acad. Sci. USA, (9): p Abraham, F.F., et al., Simulating materials failure by using up to one billion atoms and the world's fastest computer: Brittle Fracture. P. Natl. Acad. Sci. USA, (9): p Vashishta, P., R.K. Kalia, and A. Nakano, Large-scale atomistic simulations of dynamic fracture. Comp. in Science and Engrg., 1999: p Vashishta, P., R.K. Kalia, and A. Nakano, Multimillion atom molecular dynamics simulations of nanostructures on parallel computers. Journal of Nanoparticle Research, : p Kadau, K., T.C. Germann, and P.S. Lomdahl, Large-Scale Molecular-Dynamics Simulation of 19 Billion particles. Int. J. Mod. Phys. C, : p Tadmor, E.B., M. Ortiz, and R. Phillips, Quasicontinuum analysis of defects in solids. Phil. Mag. A, : p Shenoy, V.B., et al., Quasicontinuum Models of Interfacial Structure and Deformation. Phys. Rev. Lett., : p Knap, J. and M. Ortiz, An analysis of the quasicontinuum method. J. Mech. Phys. Sol., (9): p * Mainly treating multi-scale modeling related to mechanical properties

64 Some references Nakano, A., et al., Multiscale simulation of nanosystems. Comp. in Science and Engrg., 2001: p Rountree, C.L., et al., Atomistic aspects of crack propagation in brittle materials: Multimillion atom molecular dynamics simulations. Annual Rev. of Materials Research, : p Zimmerman, J.A., et al., Surface step effects on nanoindentation. Phys. Rev. Lett., (16): p Buehler, M.J., et al., Atomic Plasticity: Description and Analysis of a One-Billion Atom Simulation of Ductile Materials Failure. Comp. Meth. in Appl. Mech. and Engrg., Sharma, A., R.K. Kalia, and P. Vashishta, Large multidimensional data vizualization for materials science. Comp. in Science and Engrg., 2003: p Sharma, A., et al., Immersive and interactive exploration of billion-atom systems. Presence-teleoperators and Virtual Environments, (1): p Abraham, F.F. and H. Gao, Anamalous Brittle-Ductile Fracture Behaviors in FCC Crystals. Phil. Mag. Lett., : p Argon, A., Brittle to ductile transition in cleavage failure. Acta Metall., : p Cheung, K.S. and S. Yip, A molecular-dynamics simulation of crack tip extension: the brittle-to-ductile transition. Modelling Simul. Mater. Eng., : p Rice, J.R. and R.M. Thomson, Ductile versus brittle behavior of crystals. Phil. Mag., : p Buehler, M.J. and H. Gao, eds. Ultra large scale atomistic simulations of dynamic fracture. Handbook of Theoretical and Computational Nanotechnology, ed. W.S.a.A. Rieth. 2005, American Scientific Publishers (ASP). Cuitino, A.M., et al., A multiscale approach for modeling crystalline solids. Journal of Computer-Aided Materials Design, (2-3): p Buehler, M.J., F.F. Abraham, and H. Gao, Hyperelasticity governs dynamic fracture at a critical length scale. Nature, : p Broberg, K.B., Cracks and Fracture. 1990: Academic Press. Freund, L.B., Dynamic Fracture Mechanics. 1990: Cambridge University Press, ISBN Courtney, T.H., Mechanical behavior of materials. 1990: McGraw-Hill. Buehler, M.J., H. Gao, and Y. Huang, Continuum and Atomistic Studies of the Near-Crack Field of a rapidly propagating crack in a Harmonic Lattice. Theoretical and Applied Fracture Mechanics, : p Buehler, M.J., H. Gao, and Y. Huang, Continuum and Atomistic Studies of a Suddenly Stopping Supersonic Crack. Computational Materials Science, (3-4): p Buehler, M.J. and H. Gao, Biegen und Brechen im Supercomputer. Physik in unserer Zeit, (1). Buehler, M.J., F.F. Abraham, and H. Gao, Stress and energy flow field near a rapidly propagating mode I crack. Springer Lecture Notes in Computational Science and Engineering, ISBN : p Abraham, F.F., et al., Spanning the length scales in dynamic simulation. Computers in Physics, (6): p Abraham, F.F., et al., A Molecular Dynamics Investigation of Rapid Fracture Mechanics. J. Mech. Phys. Solids, (9): p Abraham, F.F., et al., Instability dynamics of fracture: A computer simulation investigation. Phys. Rev. Lett., (2): p Abraham, F.F., Dynamics of brittle fracture with variable elasticity. Phys. Rev. Lett., (5): p Marder, M. and S. Gross, Origin of crack tip instabilities. J. Mech. Phys. Solids, (1): p Marder, M., Molecular Dynamics of Cracks. Computing in Science and Engineering, (5): p

65 Some references Buehler, M.J. and H. Gao, Dynamical fracture instabilities due to local hyperelasticity at crack tips. under submission, Ogden, R.W., Nonlinear Elastic Deformations. 1984: Wiley and Sons, New York. Bathe, K.J., Finite Element Procedures in Engineering Analysis. 1982: Prentice-Hall. Boresi, A. and K.P. Chong, Elasticity in Engineering Mechanics. 2000: Wiley-Interscience, New York. Shenoy, V., et al., An Adaptive Methodology for Atomic Scale Mechanics - The Quasicontinuum Method. J. Mech. Phys. Sol., : p Norskov, J.K., J. Schiotz, and K.W. Jacobsen, CamposASE. Park, M. and P. Fishwick, An integrated environment blending dynamic and geometry models. ARTIFICIAL INTELLIGENCE AND SIMULATION LECTURE NOTES IN COMPUTER SCIENCE, : p Aivazis, M., The Pyre Framework Hinsen, K., The molecular modeling toolkit: A new approach to molecular simulations. Journal of Computational Chemistry, (2): p Car, R. and M. Parrinello, Unified Approach For Molecular Dynamics and Density Functional Theory. Phys. Rev. Lett., : p Duin, A.C.T.v., et al., ReaxFF: A Reactive Force Field for Hydrocarbons. J. Phys. Chem. A, : p Cheung, S., et al., ReaxFF$_\rm MgH$ Reactive Force Field for Magnesium Hydride Systems. J. Phys. Chem. A., : p Nielson, K.D., et al., Development of the ReaxFF reactive force field for describing transition metal catalyzed reactions, with application to the initial stages of the catalytic formation of carbon nanotubes. J. Phys. Chem. A., : p. 49. Chenoweth, K., et al., Simulations on the thermal decompositions of a poly(dimethylsiloxane) polymer using the ReaxFF reactive force field. J. Am. Chem. Soc., : p Duin, A.C.T.v., et al., ReaxFF$_\rm SiO$: Reactive Force Field for Silicon and Silicon Oxide Systems. J. Phys. Chem. A, : p Han, S.S., et al., Optimization and application of lithium parameters for the reactive force field, ReaxFF. Journal Of Physical Chemistry A, (20): p Chenoweth, K., et al., Simulations on the thermal decomposition of a poly(dimethylsiloxane) polymer using the ReaxFF reactive force field. Journal Of The American Chemical Society, (19): p Strachan, A., et al., Thermal decomposition of RDX from reactive molecular dynamics. Journal Of Chemical Physics, (5). Cheung, S., et al., ReaxFF(MgH) reactive force field for magnesium hydride systems. Journal Of Physical Chemistry A, (5): p van Duin, A.C.T., et al., Application of ReaxFF reactive force fields to transition metal catalyzed nanotube formation. Abstracts Of Papers Of The American Chemical Society, : p. U1031-U1031. Rappé, A.K. and W.A. Goddard, Charge eqilibration for molecular-dynamics simulations. J. of Physical Chemistry, (8): p Morse, P.M., Diatomic molecules according to the wave mechanics. II-vibrational levels. Phys. Rev., : p Stadler, J., R. Mikulla, and H.-R. Trebin, IMD: A Software Package for Molecular Dynamics Studies on Parallel Computers. Int. J. Mod. Phys. C, : p