Environment from the Molecular Level

A NERC eScience testbed project

Codes used in the eMinerals project

Large-scale molecular dynamics (MD) simulations: DL_POLY

Daresbury Laboratory is the source of the general purpose MD package DL_POLY [1] that has application across the whole range of areas to be studied under this proposal [2]. The code has parallel capability and may be run on high-end computers, Beowulf systems and single processor machines. For environmental applications, large systems of the order one million particles and beyond will be the norm. Adaptation of DL_POLY to domain decomposition parallelism [3] is essential for this. Some preliminary work has been done, but a full adaptation of the functionality of DL_POLY is awaited. In particular, adaptations for surfaces and interfaces (the Hautman-Klein Ewald sum [4]), long timescales (the hyperdynamics methods of Voter[5]) and mesoscale system (Dissipative Particle Dynamics -DPP6) are required.

1. W. Smith and T.R. Forester, J. Molec. Graphics (1996) 14, 136. (See also www.dl.ac.uk/TCS/Software/DL_POLY.)
2. W. Smith, C. Yong and P.M. Rodger, Mol. Sim. (2001) submittted.
3. W. Smith. Comput. Phys. Commun. (1992) 67, 398.
4. J. Hautman and M.L. Klein, Molec. Physics (1992) 75, 397.
5. A.F. Voter, Phys. Rev. Lett. (1997) 78, 3908.
6. R.D. Groot and P.B. Warren, J. Chem. Phys. (1997) 107 4423.

Surface dynamics

Despite the success in recent years of modelling surfaces, atomistic simulation studies that incorporate realistic descriptions of the interatomic forces have been restricted to considering charge neutral surfaces. The difficulty is that for charged systems the cluster methods do not treat the long ranged electrostatic interactions correctly and the periodic simulation cells have unphysical high concentrations of charged species. Thus there is a need to develop new a computer simulation code for overcoming these difficulties but which uses the advantages of each. This new approach will benefit by the close co-operation between those performing the science, computer scientists, for developing new algorithms and modellers in optimisation methods.

Order-N quantum mechanics (QM): SIESTA

The applicability of empirical interatomic forces is limited when the chemical environment around the atoms changes significantly. Recent progress in first-principles QM techniques based on density-functional theory (DFT) that scale linearly with system size (as opposed to the usual cube scaling) will enable us to face the challenges proposed here. One of the main pioneering projects is SIESTA, co-developed by Dr. Artacho (Cambridge). It has been successfully applied to for other types of studies with many hundreds of atoms. Environmental studies demand important developments to allow for systems with thousands of atoms. These include improved linear-scaling solvers with functionals that are more robust, greater optimisation of atomic basis sets, and improved parallelisation based on the algebra of sparse matrices (in collaboration with Julian Gale, Imperial College). The high energies for radiation damage will require incorporation of electronic excited states. These challenges will be generally transferable to other areas of e-science, and Dr Artacho is collaborating with Prof Gillan’s team at UCL in this respect.

1. Artacho E, Sanchez-Portal D, Ordejon P, Garcia A, Soler JM. Phys Stat Sol (b) 215, 809 (1999).
2. Sanchez-Portal D, Ordejon P, Artacho E, Soler JM. Int J Quantum Chem 65, 453 (1997).
3. MS Craig, MC Warren, MT Dove, JD. Gale, D Sanchez-Portal, P Ordejon, JM Soler, E Artacho. Phys Chem Min (submitted)

Embedded Cluster Methods

A long-standing problem in the atomistic modelling of minerals (and solid state science in general) is the need to combine an accurate ab initio treatment of a local region (representing for example an adsorbed molecule or a radiation-induced defect) with an approximate, but reliable treatment of the surrounding environment. The Royal Institution group, in collaboration with the Daresbury Laboratory and UCL Physics Department, is developing procedures for treating this key problem in oxide and silicate materials. The methods will make extensive use of current and future HPC facilities, and be well placed to exploit new GRID technologies.

1. Sushko PV, Shluger AL, Catlow CRA, Baetzold RC. Radiation effects and defects in solids, 151, 215 (1999).
2. Nygren MA, Gay DH, Catlow CRA. Surface Science, 380, 113 (1997).
3. Sushko PV, Shluger AL, Catlow CRA. Surface Science, 450, 153 (2000)

Monte Carlo methods

A range of Monte Carlo (MC) methods in simulation and modelling processes will be investigated, since these methods scale in a way that is very suitable for parallel and large-scale computation. The project will develop an ongoing SMART project in Reading investigating parallel stochastic methods and algorithms for large-scale problems. Parallel MC algorithms are designed based on their generic properties, and then placed in a distributed heterogeneous computing environment. This will give a generic framework in which to map MC methods into GRID environments, developing the relevant plug-ins for HARNESS and Globus tools. These methods will be developed as search tools for scanning molecular dynamics simulations, and for scaling from molecular-level simulations to larger-scale modelling.

f. Quantum Monte Carlo

Quantum Monte Carlo (QMC) is a computational method for solving Schrödinger’s equation that does not rely on the approximations inherent in current quantum mechanical techniques such as DFT and Hartree-Fock (HF). Although DFT and HF methods have dominated computational physics and chemistry over the past 20 years or so, they are limited in that electron correlations are only approximately known. Although the correlation energy is only a small part of the total energy of a system, it can be a large part of the bonding energy and thus dominate the properties of some highly correlated systems. This can be a problem for many important environmental systems, particularly transition metals (§3). QMC provides a direct treatment of quantum many body effects and, therefore, explicitly includes electron correlation. The price, however, is that the calculations are three orders of magnitude more computationally intensive than DFT or HF methods, which means that it is only rarely applied to complex problems. Although QMC is very computationally demanding, the algorithms are intrinsically parallel with minimal communication overhead as with all MC methods. This makes it ideal for distributed computing on PCs, with one PC orchestrating the whole simulation, which runs on the other PCs independently runs a simulation. This work will be carried out in collaboration with Prof Gillan (UCL).

Integration of methodologies

For the preceding methodologies, increases in accuracy are accompanied by increases in computational demands, which reduce their applicability to large sizes and long times. However, highest accuracy is not needed over the whole system and at all times: many atoms behave in simply tractable manners during longs intervals, the main subtleties being many times localised to small regions and/or short times. It is possible to take advantage of this fact by treating different regions of the system at different levels of theory. This is difficult, but has been successfully attempted in studies of epitaxial surface growth [1], nanoindentation [2], and biomimetics [3]. The development of an integrated methodology is essential for the study of pressing environmental problems in order to encompass the essential length and time scales whilst optimising the levels of accuracy require at each scale.

1. Fichthorn KA, Scheffler M. Phys Rev Lett 84, 5371 (2000).
2. Smith GS, Tadmor EB, Kaxiras E. Phys Rev Lett 84, 1260 (2000).
3. Rothlisberger U. Abstr Pap Am Chem S 219, 334 (2000)

Page is maintained by Steve Parker