Chem Grid and grid based molecular simulators Antonio

Chem. Grid and grid based molecular simulators Antonio Laganà Dept of Chemistry Univ. Perugia (I)

Summary Molecular simulators and the need for resource sharing in Metalaboratories COST Chemistry and European Metalaboratories A Grid for chemistry and molecular sciences: the GRID. IT project A Grid for chemistry and molecular sciences: the EGEE project

WHAT IS USEFUL AN A PRIORI MOLECULAR SIMULATION FOR • • • Life and biological processes and structures Innovative materials and nanodevices Atmospheric and environmental processes Food treatment and action Astro and space processes ………………….

WHAT IS NEEDED FOR AN A PRIORI MOLECULAR SIMULATION • The convergence of the necessary theoretical and computational know how. • The availability of the relevant computer programs • The availability of a sufficient computing power. • The assemblage of a suitable workflow management or problem solving environment SINCE ALL THIS IS NOT AVAILABLE IN A SINGLE SITE THERE IS NEED FOR A METALABORATORY

THE METALABORATORY • The METALABORATORY is a cluster of geographically distributed laboratories sharing expertise, hardware and software on a computing grid.

METACHEM: Metalaboratories for complex computational applications in Chemistry The D 23 Cost-Chemistry Action (launched in 1999 to enhance the European added value of national grid initiatives in Chemistry)

THE STARTING POINT OF THE METALABORATORIES • The computers of the participating laboratories exposed outside the firewall to the grid acting as a single virtual parallel machine. • The running version (not the ones under development) of the relevant codes that the participating laboratories will implement to run concurrently on the grid for the project. • The distribution software to allocate the different tasks on the most suitable machines using the public network

Metachem: Metalaboratories for Complex Computational Applications in Chemistry MURQM: Multireference Quantum Chemical Methods DIRAC: Four Component Relativistic Quantum Chemical Calculations SIMBEX: Simulation of Molecular Beam Experiments

Metachem: Metalaboratories for Complex Computational Applications in Chemistry DYSTS: Dynamics and Spectroscopy of Systems : Relevant to Environment and Applied Chemistry MURQM: Multireference Quantum Chemical Methods ELCHEM: E-learning Technologies for Chemistry ICAB: Integration of Codes for Ab Initio Methods

EU GRID for Chemistry: D 23 COST action Simbex Murqm Dirac Elchem Icab Dysts Comovit

RESEARCH GROUPS per COUNTRY (50) 1 2 3 4 5 6 9 Isr, Pl, Sk, Nl, Ch Cz, Fr, Dk, A, Sw, No Hu Gr E D, Uk, I

THE ITALIAN CHEMGRID • Design of the GRID. IT Italian computer science project and application to MIUR (Vanneschi, Pisa) • Signature of the CHEMGRID cooperative agreement • Clustering of 8 Italian University, CNR and ENEA Laboratories around CHEMGRID • Selection of a prototype molecular simulator for grid experiments

The Italian GRID project GRID. IT Enabling platforms for high performance computational Grids oriented to scalable virtual organizations CNR, INFN, CNIT, ASI, Universities

Astrofisica Bioinformatica Geofisica Chimica Computazionale Componenti HP Librerie Portali PSE Modelli di costo Sicurezza Gestione risorse GARR Osservazioni Terra Applications Program. Ming tools Comunicazioni Monitoraggio Reti ottiche Middleware High performance nets

THE WORKING GROUPS • • 1. Grid oriented Optical switching paradigms 2. High performance photonic testbed 3. Grid development 4. Security 5. Data intensive core services 6. Knowledge services 7. Portal for efficient grid access

• 8. High performance component based programming environment • 9. Grid enabled scientific libraries • 10. Grid applications for Astrophysics • 11. Grid applications for Earth observation • 12. Grid applications for biology • 13. Grid applications for molecular virtual reality • 14. Grid applications for geophysics • 15. Project management

GRID. IT Logical interaction among WPs Area 4: Applications WP 10 WP 11 WP 12 WP 13 WP 14 Area 1: Middleware and Programming Tools WP 7 WP 8 WP 4 WP 1 WP 2 Area 2: Photonic Testbeds WP 9 WP 6 WP 3 WP 5 Area 3: Grid Deployment

THE PROTOTYPE MOLECULAR SIMULATOR: SIMBEX • Managing an a priori simulation to be interfaced with the experiment in crossed molecular beam measurements Exper. Simul.

SIMBEX: a research/educa tional tool for the simulation of elementary chemical reaction • High interactivity • Advanced visualization • In deep insight into the chemical mechansm

The a priori molecular simulator Start Interaction Dynamics Observables NO Theoretical and experimental results agree? SI End

The INTERACTION module START INTERACTION Is there NO a suitable Pes? YES Import the PES routine DYNAMICS Are ab initio calculations available? YES Application using fitting programs to generate a PES routine NO Are ab initio calculations feasible? Force fieldapplication taking NO empirical data from database to generate a PES YES Ab initio application using programs for electronic structure

The DYNAMICS module DYNAMICS Are quantum dynamics calculations inappropriate? YES TRAJ: application using classical trajectory calculations OBSERVABLES NO Is the calculation single initial state? YES TD: application carrying out timedependent quantum calculations NO TI: application carrying out time-independent quantum calculations

The OBSERVABLES module INTERACTION OBSERVABLES Is the observable a state-to-state observable? Beam VM for Intensity in the Lab frame NO YES DISTRIBUTIONS: VM for scalar and vector product distributions, and state-to-state crosssections NO YES END Do calculated and measured properties agree? Is the observable a state specific observable? YES CROSS: VM for state specific cross sections, rate constants and maps of product intensity NO RATE: virtual monitor (VM) for thermal rate coefficients

THE ANGULAR DISTRIBUTION VIRTUAL MONITORS FOR ATOM DIATOMS H+ICl→H+ICl→HCl+I H+ICl→HI+Cl

Partner Laboratories • INFORMATICS: Vanneschi (Pisa), Gervasi, Tasso (Perugia), Kacsuc (Budapest), Tirado (Madrid) • CHEMISTRY: Laganà, Tarantelli, Rosi (Perugia), Vittadini (Padova), Barone (Napoli), Capitelli (Bari), Rosato (Roma), Casalone (Milano), Zannoni (Bologna), Alberti (Barcelona), Garcia (Vitoria), Balint-Kurti (Bristol), Parker (Norman), Lendvay (Budapest), Clary (Oxford), Nyman (Goteborg)

Chem. Grid. it: a Grid model for Chemistry MI CILEA RM PD UPV PG UB CESCA NA BO BA CINECA

Chem. Grid: the PG configuration Access to GRID. IT

COLLABORATIVE USE Perform extended computational campaigns for systems relevant to scientific and technological applications Develop grid tools: middleware, workflow managers, problem solving environments and coordination languages for distributed heterogeneous environments Specialize in some specific application

EXTENDED COMPUTATIONAL CAMPAIGNS FOR FEW ATOM SYSTEMS • Estimate detailed bimolecular collision properties (such as vector correlations like k*k’) using both quantum, semiclassical and quasiclassical means (especially for scarcely reacting systems) • Investigate collision mechanisms of startup or rare events • On the fly evaluation of internal energy distributions, cross sections, rate coefficients for extended intervals of energy and temperatures in non equilibrium systems (like shock waves, plasmas, etc. )

The molecular dynamics problem Separation of electronic and nuclear motions Electronic Schrödinger equation: Nuclear Schrödinger equation:

EQUATIONS OF MOTION • Classical: integrate in time the classical mechanics equations for particles • Quantum - (time independent): separate time and integrate over the reaction coordinate after expanding over a proper basis set and inteegrating over bound coordinates - (time dependent or wavepacket): integrate over time the system wavepacket set at time 0 at the reactant asymptote

DISTRIBUTION TOOLS MPI fine and coarse grain concurrent treatment Skeleton based concurrent treatment

The implementation using MPI Inserting into the text directives from library routines • SEND • RECEIVE • BROADCAST • BARRIER

MPI TRAJECTORY PSEUDOCODE Master: DO traj_index =1, traj_number RECEIVE status message IF worker “ready” THEN generate seed SEND seed to worker ELSE GOTO RECEIVE end. IF end. DO Worker: SEND “ready” status message RECEIVE seed integrate trajectory update indicators SEND “ready” status message GOTO RECEIVE

RWAVEPR: wavepacket scalar pseudocode Read input data: v, j, k, masses… Perform preliminary calculations Loop on J Loop on t Loop on Λ Perform time-step propagation Perform the asympotic analysis Calculate C(t) coefficients and update the fixed-J S matrix End loop on Λ End loop on t End loop on J

FFT time step propagation pseudocode Loop on R Perform the Fourier transform on the R row Multiply by the R momentum Perform the back Fourier transform End loop on R Loop on r Perform the Fourier transform on the r column Multiply by the r momentum Perform the back Fourier transform End loop on r Loop on R Loop on r Multiply wavepacket elements times V elements End loop on r End loop on R

DVR-2 D time step propagation pseudocode Loop on R Loop on r Multiply the A rows by the column of C End loop on r End loop on R Loop on r Multiply the B rows by the C row End loop on r End loop on R Loop on r Multiply the V element by the C element End loop on r End loop on R

Fine grain parallel model: DVR Worker Master Do i=1, N Compute Row(i, HPS) End do

Skeleton based Coordination languages ü Sequence ü Task farm ü Loop ü Pipe

The generalized Parmod module (composition and shared objects) {External objects • Global variables • Shared memory • CORBA Input stream M 1 Exit stream s 13 s 34 M 3 M 2 s 23 M 4 s 45 s 25 Parallel (or sequential) module s 54 M 5

Fine grain DVR: MPI vs ASSIST

COARSE GRAIN PARALLELISM ü SPMD - Fixed initial conditions (v, j) distributed on the nodes of the Grid ü TASK FARM Distribute jobs for fixed J (partial waves) and Λ pairs with complementary J values on the same processor of the Grid node

Coarse grain Master-worker parallel Model master slave J=0, N J=0 Λ=0 master J=N Λ=1 … J=N Λ=N Collect master slave J=N Λ=0 J=1 Λ=1 J=N-1 Λ=0 … J=N-1 Λ=N-1 Collect … END

Master-slave performances

DYNAMIC AND THERMODYNAMIC PROPERTIES OF LARGE SYSTEMS • Use simplified interactions for an ensemble of millions of particles (gaseous and condensed phases) • Carry out calculations for long times • Identify mechanisms, phases and collective motions and shapes • Work out algorithms and methods for more efficient distributed computing

The intramolecular force field of a benzene nanotube Benzene intramolecular interactions Amber parameterization Cornel et al. JACS 117, 5179 (1995)

Interaction representation q Additivity of simple models q Portability

MOLECULAR APPROACHES TO VIRTUAL REALITY REPRESENTATIONS • Build remotely accessible virtual environments for research, education, training combining meter and nanometer levels • Combine distributed molecular simulations engines with local user virtual environment applications • Semantic web applications to chemical knowledge

Virt. CHEMLab: abstract model of a real Lab A laboratoy bench realized in VRML A fume-hood laboratory bench realized in VRML

Virt. CHEMLab: Flame Spectroscopy Go to VRML animation

Grid-learning services Students Teachers’ net SERVER

XML Language Name Enabling technology for Multimedia and Virtual Reality HTML CML Math. M L Javascript &VBscript Perl ASP Flash Director Macromedia VRML X 3 D Language Type Markup Object programmi ng Script Script Program ming Executed on Browser Server Browser Need compiling No Yes No Interpreter No No No Viewable Source Document Type Animation Capability Yes No No Static Active Dynam ic Active No Yes No No No Yes Yes Inserted in Web pages through <applet> Embedded in web pages through <script> Called from Web page Inserte d within <% …%> Embed ded in web pages <object > Embedde d in web pages through <object> Basic language Web for Implemetation layout

DEVELOPMENT OF GRID TOOLS • Middleware (cost algorithms, chemical knowledge representation, privacy preserving data analysis, …) • Workflow manager and problem solving environments (virtual experiments, virtual laboratories, . . ) • Distribution tools for scientific applications

The GILDA testbed sites • The Chemistry Department of the University of Perugia has been included in the sponsor and the testbed site list • The Chemistry Department node is made by a cluster of 14 nodes (2 proc. Intel Pentium III, 2 G RAM, 40 G HD) + Computing Element + LCNGFG server

The outcomes of Comp. Chem • The VO Comp. Chem has been created to register to the VO: http: //grid-cnaf. infn. it/index. php? voregister&type=1 • The implementation of the Grid Molecular Simulator prototype has prompted a modification of the EGEE infrastructure in order to guarantee the strong need for real-time interaction with the Grid • The prototype will be rewritten in XML (instead of PHP) to be included in the Genius application testbed to demonstrate the power of the Grid. • We attended the following EGEE events: – EGEE NA 4 preliminary workshop, December 15, 2003 Paris – EGEE Generic Applications Advisory Panel (EGAAP), First Meeting, Geneva, June 14, 2004 – EGEE NA 4 Open Meeting, Catania, July 14 -16, 2004 – EGEE NA 4 signature of Mo. U, July 31, 2004

Grid Molecular Simulator and GILDA • The implementation of the Grid Molecular Simulator prototype has modified EGEE infrastructure in order to guarantee the strong need of real-time interaction with the Grid • The prototype will be rewritten in XML (instead of PHP) to be included in the Genius application testbed to demonstrate the power of the Grid.

Grid Molecular Simulator on EGEE • The aim of EGEE project is to build on recent advances in grid technology and develop a service grid infrastructure in Europe which is available to scientists 24 hours-a-day. • A prototype implementation of a Grid Molecular Simulator has been selected for NA 4 Activity of EGEE: Application Identification and Support

Grid NA 4 Activities of NA 4 focus on the identification of grid earlyusers and established applications to support their implementation on the EGEE infrastructure. In particular NA 4 has the following objectives: • identify through the dissemination partners and a well defined integration process a portfolio of early user applications from a broad range of sectors (universities, industries and companies). • To support development and production use of the applications on the EGEE infrastructure to establish a strong community.