Objectives and organization

Funding for the new flagship project on Car-Parrinello in Chemistry was approved by the EPSRC in december 2000 together with the renewal of the CCP1 project. The new flagship project continues the trend towards larger and more complex systems, which was already a central topic in the previous project (1997-2000) on Hybrid methods. Similar to this project, which was based at the University of Manchester, the new project will be headed by one of the CCP1 sites, namely Cambridge. The CCP1 members involved are M. Sprik(MS), A. Alavi(AA). There is also money to appoint a PDRA for a three year period. Similar to the PDRA associated with the previous flagship project (Richard Hall) the new PDRA will be based at the coordinating institution (Cambridge). The aim as formulated in the proposal is twofold: Introduction of the CCP1 community to, and familiarization with, the Car-Parrinello technique in a number of collaborative applications to chemical problems in the condensed phase and, secondly, development of the Car-Parrinello methodology for the study of optical spectroscopy (IR, Raman and UV) of condensed molecular systems. The first part of the project intends to bring together a number of interested CCP1 members in a collaboration with the Cambridge group (MS, AA and PDRA). A number of topics and collaborations for these subprojects has already been identified focusing in surface science, chemical reactions and catalysis in solution, and the interpretation of vibrational spectroscopy (IR, Raman) probing conformational structure of organic and (models) of biomolecules. A further project on electronic spectroscopy, to be carried out in Cambridge, is more oriented towards method development.

The Car-Parrinello method and CCP1

The ab initio Molecular Dynamics (MD) technique (``Car-Parrinello'') celebrated last year its 15-th birthday (at a meeting in Minneapolis organized by the University of Minnesota Supercomputing Institute). In those fifteen years the method has undergone an evolution stretching from its roots in solid state physics to the recent application to complex molecular systems in biology. The ab initio MD method, as originally formulated by Roberto Car and Michele Parrinello [1] is a combination of solid state electronic structure calculation and molecular dynamics simulation methods (reflecting the different background of its two founding fathers). The electronic structure calculation is based on Density Functional Theory(DFT). The MD component, on the other hand, was inspired by force field based numerical simulation schemes as applied in the classical physical chemistry of liquids. As such the Car-Parrinello is the first (and possibly only) numerical method introducing finite temperature statistical mechanics in electronic structure calculation. This makes it the method of choice for the numerical study of the chemistry of condensed molecular systems (solutions, bio-polymers) where thermal fluctuations play a role (for a thorough recent review see Ref. 2).

Of course, a vital condition for the success of the extension of Car-Parrinello to chemistry is a reliable description of structure and energetics of molecules. This requires a consistent performance over at least two orders of magnitude in energy, from formation energies for chemical bonds in the eV range to the much weaker intermolecular interactions (in particular hydrogen bonding). Applications to chemistry can therefore be more demanding than the solid state physics problems for which DFT was originally invented in the sixties by Walter Kohn. Thus crucial to application of DFT to chemistry, is a thorough understanding of what ``in practice'' distinguishes the DFT of molecules from the DFT for solids. This development, initiated by Becke and others, led eventually to the acceptance of DFT by the quantum chemistry community as a tool for the computation of electronic structure of large systems (although pockets of skepsis remain).

The CCP1 community has played a leading role in this process right from the beginning. Following the flagship project 1994-1997 which was devoted to DFT, the GAMESS-UK package, developed and maintained at Daresbury with support of CCP1 was extended with DFT modules, as were several of the group codes of CCP1 members, such as CADPAC(Cambridge) and MOLPRO (Birmingham). These codes have been used to study a large variety of chemical problems, ranging from structure and reactivity to spectroscopic properties. Of special relevance in the context of the new flagship project are the contributions by the Handy group (Cambridge) on the development ofaccurate Generalized Gradient Approximations (GGA's) that can be used in plane wave codes. Continuing along the lines set out by Becke this group has produced a family of GGA's based on fits to an increasingly larger set of high-accuracy quantum chemistry and experimental data. Accurate GGA's are crucial for chemical applications of the Car-Parrinello method since plane wave basis sets make the use of hybrid functionals, such as B3LYP, prohibitively expensive.

Active involvement with the Car-Parrinello method, i.e. direct confrontation with the implications of a condensed phase environment (e.g. solvent) treated at the same level of electronic detail as the reactants, is, therefore, a natural next step for the CCP1. Very similar considerations had already prompted the theory group at Cambridge to expand its activity into this direction with the appointment of Michiel Sprik and Ali Alavi. Both have an extensive background in Car-Parrinello studies of condensed systems, with a focus on chemical reactions in solution (MS) and surface chemistry (AA). With this technical expertise available, Cambridge is also an obvious site to host a project of this kind. A first collaboration project has also already been set up. This is with the Manchester group of Prof. Hillier who want to use Car-Parrinello to look at pericyclic reactions in aqueous solution and physi- or chemi-sorption from solution onto mineral surfaces.

CPMD code

The software platform for the project is the CPMD package [3]. This is a versatile ab initio MD code which has already been used for numerous chemical applications. The code has been run and tested on virtually every commercially available hardware platform, from PC's to supercomputers. Parallellization of CPMD is implemented using message passing (MPI) and, as a result, the code is most efficient for distributed architectures, such the T3E and SP3. Moreover, the gains of parallellization on the more cost effective Beowulf clusters of PC's are also substantial. Benchmarks on the Beowulf cluster in Daresbury laboratory for a typical problem (a periodic sample of liquid water) ran at 70 percent of the CPU time needed for the same system on the T3E at CSAR for processor numbers up to 32. This favourable performance was the reason to choose the Daresbury Beowulf cluster, or future upgrades, installed and maintained by Martyn Guest and his team as the main computational resource for the special project on spectroscopy (see below). Additional equipment money of 35k pounds has awarded by the EPSRC for this purpose.

The CPMD package (in the from of the source code) is available free of charge to every CCP1 member. The only requirement for individual groups who would like participate in the collaboration or would want to make use of it for unrelated projects is signing a license agreement with Prof.~Parrinello. There are, however, no further restrictions on the distribution of CPMD, geographically or otherwise. This is of increasing importance in view of the international mix of the current academic population in the UK. In particular, this means that a postdoctoral fellow from abroad, who has been working with or on the code, is allowed to take his work with him when he moves to the next job or returns home and to continue there. This, in fact, is a distinct practical advantage of CPMD over CASTEP, the Car-Parrinello Code most popular in the UK (CASTEP is made available to UK universities through the UKCP consortium under more restrictive conditions). The main motivation of preferring CPMD over CASTEP, however, is that the experience of Sprik and Alavi, who are coordinating the project, is almost entirely based on CPMD.

Special project on condensed phase spectroscopy

Spectroscopy in its various forms (NMR, IR, Raman, UV) is vital in chemistry as an experimental technique for determination of structure and dynamics of molecules. Computation of molecular spectra has also a long tradition in quantum chemistry and has stimulated many important methodological advances. For example, accurate computation of vibrational spectra and dielectric response crucially depended on the development of analytical derivative techniques. Application of these methods to NMR (chemical shifts) gave rise to further development in treatment of magnetic effects. Also the current rapid development of time dependent density functional methods for the calculation of electronic excitation spectra can be seen as an illustration of the interaction between spectroscopy and computational chemistry.

Spectroscopy, until recently, has not received comparable attention in Car-Parrinello based research. One of the reasons for this neglect was the infamous ambiguity in the specification of polarization in periodic models of condensed phase systems. Fortunately, a couple of years ago, these problems could be finally resolved by Vanderbilt and coworkers [4]. As shown by Resta [5] the Vanderbilt approach ultimately amounts to a proper definition of the position operator in periodic cells. The Vanderbilt method has been implemented in CPMD and used in a first principle estimate of the molecular dipole moment [6] and IR absorption in liquid water [7]. A further major step forward was made with the implementation of a variational perturbation scheme for computation of analytical second derivatives in plane wave basis sets [8]. This opens the way to computation of both vibrational and NMR spectra.

These technical developments are only in an initial phase and a number of problems remain to be resolved. They hold, however, the promise of providing us eventually with a technique for parameter free computation of optical and NMR spectra of complex molecular systems in the condensed phase. This could have a major impact on the structural interpretation of experimental spectra of organic molecules in solution (particular aqueous solution) or other strongly interacting environments. The Cambridge group will join this effort by focusing on vibrational spectra. By linking this with the CCP1 project we hope to able to make rapid progress using the extensive experience with molecular spectroscopy gathered over the years by quantum chemists. One of the more ambitious objectives of this project is to be able to predict the effect of conformation of biomolecules on their IR spectra in collaboration with Dr Reynolds (Essex).

The Car-Parrinello community is also trying to catch up with the computation of electronic spectra using time dependent density functional methods. This development has only just started with a verification that plane wave - pseudo potential based calculations are indeed capable of reproducing benchmark results obtained by atomic basis set techniques, such as the UV spectra of the nitrogen molecule, formaldehyde and formamide [9]. Extending these calculations to molecules in solution is a major challenge which will also be taken up by the Cambridge group as part of the flagship project. The problems ahead, however, are substantial, both in terms of computational methodology, but also at the more fundamental level of the DFT of coupled electronic excitations in molecular systems. Therefore, the emphasis in this project will be more on development of method and code. Validation of results and the resolution of the more fundamental issue concerning DFT will require continuous interaction with experienced quantum chemist (such as for example the Handy group at Cambridge) using different techniques, DFT or wavefunction based. CCP1 is a uniquely suitable forum for development of such novel methods involving many aspects of quantum chemistry.

References

1. R. Car and M. Parrinello, Phys. Rev. Lett. 55 , 2471 (1985).
2. D. Marx and J. Hutter, J, Ab initio Molecular Dynamics: Theory and Implementation in Modern Methods and Algorithms of Quantum Chemistry, J. Grotendorst (Ed.), Forschungszentrum Jülich, NIC Series, Vol 1, pp. 301-449 (2000).
3. CPMD, developed by J. Hutter, A. Alavi, T. Deutsch, M. Bernasconi, S. Goedecker, D. Marx, M. Tuckerman, M. Parrinello at the MPI für Festkörperforschung (Stuttgart) and the IBM Zürich Research Laboratory 1995-2000.
4. R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651 (1993).
5. R. Resta, Phys. Rev. Lett. 80, 1800 (1998).
6. P. L. Silvestrelli and M. Parrinello, J. Chem. Phys. 111, 3572 (1999).
7. P. L. Silvestrelli, M. Bernasconi and M. Parrinello, Chem. Phys. Lett. 277, 478 (1997).
8. A. Putrino, D. Sebastiani, and M. Parrinello, J. Chem. Phys. 113, 7102 (2000).
9. N. L. Doltsinis and M. Sprik, Chem. Phys. Lett. 330, 563 (2000).