Large-scale molecular dynamics (MD) simulations: DL_POLY3

DL_POLY_3 is a general purpose MD package designed for large-scale molecular dynamics (MD) simulations. It delivers ultimate parallelism to any high processor count (m) applicable to the problem size (N ≤ 2.1 billion particles) with excellent performance of order N and NlogN for short and long range interactions respectively. The code is fully self-contained and written in a high-quality, modularised FORTRAN90 (+MPI) which guarantees exceptional portability. The package functionality is fully documented in a searchable and internally self-referencing PDF manual.

The parallelisation design is based on spatial domain decomposition (DD) which provides inherent parallelism - indifferent to processor count (m) and topology, and full memory distribution of the problem (N/m). This guarantees quasi-equal workload balancing provided the problem's density distribution is quasi-uniform. The DD of the MD system is complemented by further partitioning of each domain into linked cells (LC) to facilitate the rapid evaluation of short range interactions. The evaluation of the long range contributions of the Coulomb interactions is based on the Smoothed Particle Mesh Ewald (SPME) method employing three dimensional fast Fourier transform (3D FFT) operations which respect and benefit from the DD and LC partitionings.

Integration is available in both leapfrog Verlet (LV) and Trotter derived velocity Verlet (VV) couched flavours. The latter is default as it provides symplecticity (time-reversibility), better numerical stability and exact estimators. Integration is can be chosen to produce any important thermodynamic ensemble (NVE,NVT,NPT,NsT) in a specific implementation if available (Nose-Hoover, Berendsen, Martyna-Tuckerman-Klein).

The DL_POLY_3 package is also available as a CML enabled code with the integrated eMinerals technologies AgentX [link here] and FoX [link here] for reading and printing CML respectively. This allowes for certain inter-operability between this and other CML enabled codes (such as SIESTA [link here] and GULP [link here]) by shedding away the specificity of the different I/O textual formats for different applications. For example, the CML marked up output of a quantum mechanically (QM) optimised molecular structure output by SIESTA can be directly fed in DL_POLY_3 and simulated. CML enabled output can be XSLT processed to DHTML and then browsed in real time while the simulation is running (this includes 2D plots of various quantities versus time and of the simulated molecular structure provided Adobe SVG plugin is available). Also, graphical packages such as Atomeye and Eutopia can read and display CML marked up molecular configurations without the need of parsers. Furthermore, the programme can be integrated to run on a computer grid (via Globus [link here]) with I/O connected to a data grid (SRB [link here]) and thus do a simulation sweep simultaneously on many processors (i) over various chemically related compounds at the same thermodynamic conditions or (ii) over various thermodynamic conditions of the same compound. When the sweep is done the data grid can be asked using AgentX and a scripting language to produce tables of how properties vary over the range of the sweep which can be visualised using XSLT script.

The qualities of DL_POLY_3 make it a particularly irreplaceable tool in large scale simulations (where the size of the problem is beyond hundreds of thousands) and in simulations of highly non-equilibrium processes. The package has various features that help deal more efficiently with (variable time step, density variation) and give deeper insight (defect detection, particle displacements, radial distribution functions) in such simulations. Radioactive waste encapsulation and change in properties of materials exposed [link to highlights] to irradiation are the most relevant environmental issues particularly suited to be studied with the use of DL_POLY_3.

DL_POLY_3 has set the record for largest system simulated by a general MD package. In March 2006 it ran a simulation of a 300 million atom system of NaCl with full Ewald sumation of the coulombic interactions on 1024 CPUs with 2GB/CPU RAM of the HPCx cluster.

General references

Papers that describe the DL_POLY can be downloaded as pdf files from the following links:

---