SIMBEX: A METALABORATORY FOR THE A PRIORI SIMULATION OF CROSSED MOLECULAR BEAMS EXPERIMENTS

COST ACTION D23

COST Program # D23/0003/01

Start Date: 01/02/2001     Expiration Date: 18/07/2005

APPLICANTS

Short description of the proposal

The project is aimed at constructing a simulator of molecular beam experiments operating through the Web. This will be achieved by assembling the expertise of several chemical laboratories competent in carrying out electronic structure and dynamics calculations and skilled in running molecular beam experiments. Computer science groups will also collaborate to construct the simulation on the Web, to develop related middle-ware and to manage the metacomputing back-end. The simulator will be used to reproduce reactive scattering properties of some gas phase chemical reactions, to rationalize their attack mechanisms and to pivot experimental measurements.

Keywords: Molecular simulations, problem solving environments, metacomputing, metalaboratories.

Research proposal

The simulation of molecular beam measurements is a computing application in which one has to give both a detailed a priori description of the molecular nature of the process being investigated and a statistical and a graphical treatment of calculated data to reproduce the experimental situation. In this case a Metalaboratory environment allows the investigators to use on the network highly complex packages to calculate electronic energies, to integrate nuclear dynamics equations, to collect the necessary information from specialistic data banks, to perform the necessary statistical manipulations as well as to render the results using graphics, animation and virtual reality techniques.

Appropriate tools and environments are being developed to allow scientists to use distributed computing platforms of a computational grid without caring about related technical aspects and eventually making use of the language of the specific scientific field. There is a growing number of worldwide projects concerned with the investigation of these aspects[1,2]. Some of them focus on the exploration of technologies for Web based metacomputing (a review of related tools is available [3,4]). Among them is Charlotte[5], developed at the New York University, that is the first environment to allow Web machines to take part into an ongoing computation without having to rely on a native code. Javelin[6], developed at the University of California in Santa Barbara, is also a Java based infrastructure for global computing exploiting the potentialities of Internet and Web based technologies. Webflow[7] developed at the Northeast Parallel Architecture Center is a computational extension of the Web model acting as a a framework for the wide area distributed computing and metacomputing. Netsolve[8], developed at the University of Tennesee and the Oak Ridge National Laboratory, is a client-server application designed to solve computational science problems using a distributed computing environment.

Software products specifically designed for distributing computations on clustered computers are: Condor[9] (this software implements a high performance computing environment allowing the exploitation of the computing power of a cluster of UNIX workstations by delivering submitted jobs to the most suited machine and redirect them to another machine when the owner starts a job session), LoadLeveler[] (this software makes use of job schedulation, checkpoint and features similar to those of Condor. LoadLeveler provides the user with a graphical interface for sumitting requests to run sequential and parallel jobs implemented using PVM, MPI or MPL), Codine[11] (this software, presently marketed by SUN as GridEngine, optimizes the use of hardware and software resources within a heterogeneous computing environment such as clusters of workstations and clusters of vector and parallel supercomputers through friendly interface), LSF[12] or Load Sharing Facility (this is a product that allows the execution of sequential and parallel applications either in interactive or in batch mode and makes use for parallel execution of PVM, MPI or P4 message passing libraries). These products are referred to as Distributed Resource Management (DRM) software. The New Productivity Initiative, see http://www.newproductivity.org, is seeking to produce a recommendation for an open interface standard which will enable inter-working of DRM software and applications in a portable way. More recently grid software like Globus, StaMPI, Pacx-MPI, MPICH-G has also been produced with the purpose of enabling multi-institutional research efforts to provide a high performance worl wide computing environment for complex computation oriented applications[13].

Some of the applicants are already collaborating in a COST working group to develop computer codes aimed at calculating potential energy surfaces, integrating differential equations for evaluating reactive properties of elementary systems, performing the necessary statistics to simulate quantities measured quantities. This collaborative work has been developed within the activity of the COST in Chemistry Action D9. The present proposal aims at using the outcome of this collaboration to construct a Web based simulator of crossed molecular beam experiments (SIMBEX) by assembling an ad hoc Metalaboratory environment.

Ab initio calculations of potential energy surfaces are usually performed using customized variants of established quantum chemistry computational packages. Among these are NWChem[14], Gaussian[15], MOLPRO[16], CADPAC[17], Turbomole[18], HONDO[19], etc.. The code considered for SIMBEX is GAMESS[20].

Reactive scattering calculations are usually performed using computer codes internally developed by the various laboratories and none of these has yet reached the stage of commercialization (as it has happened for electronic energy codes. Investigations to develop reactive scattering codes are being carried out by several groups in the world. Among these A. Kuppermann (Pasadena), D. Truhlar (Minneapolis), M. Baer (Yavne), J. Bowman (Atlanta), E. Goldfield (Detroit), D. Manolopoulos (Oxford) J. Light (Chicago), G. Parker (Norman) and J. Zhang (New York). Some of these codes are presently available for distribution by the Quantum Chemistry Program Exchange Library like VENUS96 of W.H. Hase[21] and DYNASOL of J.Z.H. Zhang[31].

Applicants' research

The first subgroup is articulated as follows:

The Perugia group, that is also the principal applicant, has a particular competence in designing new theoretical and computational approaches for accurate full and reduced dimension quantum calculations of reactive properties of atom-diatom and diatom-diatom systems. The group has also gained expertise on dealing with the desig and implementation of parallel models to distribute reactive scattering calculations on massively parallel architectures. The group is also active in carrying out molecular beam experiments.

The Vitoria group has competences in constructing potential energy surfaces suitable for dynamical calculations and in performing classical trajectory calculations and reduced dimensionality time independent quantum calculations for systems made of three and four atoms. The group has also the know how for extending these calculations to model treatments of more complex systems and for generalizing the potential energy fitting based on the bond order coordinates developed originally for three atoms to four and more atoms.

The Bristol group interests cover several areas in molecular dynamics, e.g. photodissociation, inelastic scattering and reactive scattering processes. Its competence extends over the molecular electronic structure computations and the time-dependent quantum reactive scattering calculations. In this particular area, they have recently developed a new method, which is much simpler and more efficient than any previously available one.

The Goteborg group focuses its research on the quantum dynamics of chemical reactions involving polyatomic molecules with four, six and even seven atoms. This typically only allows a numerical exact treatment of a reduced number of coordinates (while other coordinates, less important for reaction, must be treated in an approximate way e.g. adiabatically). The adoption of a reduced dimensionality approach simplifies the construction of a realistic potential energy surface for the selected degrees of freedom and makes the extension of the dynamical calculations to large polyatomic molecules feasible.

The Budapest group has a well-known experience in ab initio calculations of potential energy values for large molecules and in related classical trajectory evaluations of dynamic and kinetic properties of complex systems. A particular skill of this group is its ability to find relationships between features of the potential energy surface and the characteristics of the calculated reactive properties.

The second subgroup is articulated as follows:

The Pisa group is particularly skilled in dealing with parallel and distributed computational applications. They have also developed problem solving environments and software for managing applictions running on geographically distributed platforms.

The Madrid group is the main department associated with the Supercomputer Centre of the Complutense University in Madrid. In the last few years, this group has developed several parallel applications and problem solving environments for simulating a large variety of physical phenomena in computational fluid dynamics, weather prediction and non-linear optics. Its research interests include the effective exploitation of high performance parallel computing systems and the development of user-friendly environments that help parallel computer users to employ such systems.

The SZTAKI group is a Centre of Excellence recognized by the EU. It has a laboratory dealing with supercomputing, cluster computing and grid computing. The group develops a graphical parallel programming environment (including compiler, mapping, debugger, monitoring and visualization tools) for end-users coming from the field of various science branches like chemistry and biology. The group is also active in developing performance and monitoring tools for grid middle-ware.

The Poznan group is a Supercomputing and Networking Center of Poland and is affiliated to the Intitute of Bioorganic Chemistry of the Polish Academy of Sciences. The Center consists of 5 departments. The Application Department works very closely with various application developers, especially those from computational chemistry. The department includes also a strong group working on the tools and systems for metacomputing. The most important projects are: Multicriteria Resource and Service Brokers for Grids (MC-Broker), Computational Grid Portals, (eg. Bioinformatics Portal). The group is specialised in reource management in grid environments. The center is also a National Research and Education Network Operator (POL-34/155). It is very experienced in networking and testbeds.

The Daresbury group is part of the Central Laboratory of the Research Councils (CLRC) is the leading UK High Performance Computing Group supporting 70% of the applications on national suprtcomputers via its HPCI Centre, undertaking investigations into the use of new computing technology ans software via the UKHEC Collaboration and evaluating and deploying Computational Grid technology as part of the CLRC e-Science Centre. Particular areas of expertise are in computational chemistry and materials and in numerical algorithm design and performance optimisation on distributed systems. The CLRC is a High Performance Computing Center of UK that has a special section taking care of developing computer codes relevant to chemical applications and implementing their parallel and distributed versions. The group develops also software for the measurement of the performances of programs running on concurrent processor architectures and for their friendly usage.

The work to be carried out

The main aim of the proposed Metachem working group is twofold:

1. extend the organizational model of the PSE developed for SIMBEX to a European Metalaboratory;
2. develop Web tools for enhancing cooperative chemical simulations and activating feedbacks between a priori modeling and on-line devices.

The driving idea of SIMBEX is the development of Internet and Web-based parallel computing technologies to combine a priori modelling with experimental measurements. This implies the exploitation of the Web as an infrastructure for running coarse-grained distributed complex parallel applications and its use as a pervasive grid infrastructure [25,26] {\em i.e.} an easy to scale-up metacomputer to develop parallel and collaborative work. This also implies the development of tools for the location allocation and management of the resources, for the implementation of fault-tolerance, of security and access control, for achieving scalability, flexibility and performance enhancement. Morevoer, all this has to be implemented in a completely transparent way.

This also implies the assemblage of a chemical computing machinery that drives the user(s) from first principles to the measured signal by taking as a case study molecular beam experiments. The logic flow associated with SIMBEX (see the figure) is articulated into three blocks. This means that the partipating laboratories will have to construct a computational procedure able to: (1) provide a suitable potential energy surface based on ab initio potential energies; (2) perform dynamical calculations of the efficiency parameters of the chemical reaction being investigated; (3) assemble calculated information in a way that simulates experimental conditions and render it using graphical emulating the measured observable. More in detail this implies:

1. the generation (if not available from a data bank, locally or on the Web) of a potential energy routine through the interpolation of a sufficiently large set of potential energy values using a suitable functional  [27,28,29,30]. If ab initio data are not available, related calculations are performed using a suitable computational procedure. When this is not feasible (for economic or scientific reasons) a semiempirical computational procedure will be activated. For systems made of many atoms the potential energy surface is built by summing up two, three and four body model interactions.

2. the run of dynamics calculations using either classical or quantum approaches. In quantum approaches, then, either a time dependent technique or a time independent one can be chosen. Both time dependent and time independent quantum programs supply the value of detailed S matrix elements for a certain number of energies and quantum states. In certain cases, to make the calculations feasible, some dynamical constraints are incorporated into the dynamical treatment. This implies that the outcome of the calculations is an approximate S matrix averaged over some quantum states. For large systems the complexity of the calculation is reduced by borrowing (at least for some degrees of freedom) the simpler formalism of classical mechanics leading to quantum-quasiclassical procedures. For very large systems, one has to rely either on pure quasiclassical or statistical computational procedures. In this case one obtains directly the probability P rather than the S matrix.

3. the manipulation of S and P matrices to estimate reactive properties and build a virtual monitor emulating that of the experiment. From this the extent of confidence assignable to the simulation can be evaluated. Then the simulation can be used to rationalize the behaviour of the system, to understand the molecular mechanism driving the process and to pivot the experiment.

The work programme

1. 2001 - 2002 Implementation of a revised and extended small scale prototype on a stand alone machine. The revision with respect to the version described in ref. [23], will focus on the software used for graphical interfaces. The extension will consist in the implementation of the attachment of the package for the calculation of ab initio potential energy values and fitting routines for the first block and in the assemblage of some routines for the averaging aimed at reproducing the experimental signal.
2. 2003 Mid term workshop to evaluate the resulting product and to identify the computing resources of the cluster suitable for implementing its different parts (to be run eventually concurrently on some of them). The logic of the problem solving environment will also be discussed to incorporate the new options and develop new tools. To this end the possibility of opening the working group to other laboratories will be examined and eventually the possibility of splitting the group into two separate computer science and a chemistry group.
3. 2004 - 2005 Development of the prototype on a large and distributed scale. To this end the grid system will be defined and implemented. Related software will be tested and implemented. As a study case one or more crossed molecular beam experiments will be investigated and the potentialities of the simulation for pivoting a real crossed molecular beam machines will be exploited.
4. 2005 Final workshop and diffusion of the results.
   

Objectives and expected achievements

A second central target of the project is the development of a three level PSE articulated in an application layer, a middle-ware layer and a back-end. This will allow the construction of innovative prototypes for implementing complex chemical simulations which can provide insights on the molecular nature of chemical processes and a tool for pivoting the experiment.

• Bibliography
• About this document ...