Feature articleNWChem: A comprehensive and scalable open-source solution for large scale molecular simulations☆
Introduction
Computational modeling has become an integral part of many research efforts in key application areas in chemical, physical, and biological sciences. The ability to provide a fine level of detail through the use of electronic structure and the freedom to utilize diverse external constraints and conditions makes computational modeling an extremely versatile tool to probe various physical observables of the system. In many cases the information gained from computational studies is unique and would be difficult to obtain from experiment.
To fully realize the potential of computational modeling in answering key scientific questions requires the availability of high quality computational codes that offer wide range of capabilities and are able to take advantage of massively parallel computer architectures. One particular example of such a code is the NorthWest computational Chemistry (NWChem) software package developed in the W.R. Wiley Environmental Molecular Sciences Laboratory (EMSL) at the Pacific Northwest National Laboratory (PNNL). The NWChem software package is currently distributed to over 2800 sites world wide and has played a key role in solving a wide range of complex scientific problems. The success of NWChem stems from the fact that it offers a broad array of molecular modeling capabilities that can be deployed on all the major supercomputing platforms and that the software is freely available to the general scientific community. The latest release, version 6.0, marks a new phase in the NWChem development efforts – a transition to the open source software model under the Educational Community License.
This work presents a high level overview of NWChem focusing on the core computational modules provided by the code.
Section snippets
Parallel infrastructure
The parallel infrastructure of NWChem is built upon the Global Array (GA) toolkit [1], [2] that has been co-developed with NWChem to meet the requirements of distributed data algorithms in the computational chemistry area. The GA toolkit combines the best features of both the shared and distributed memory programming models [1], [2]. It implements a shared-memory programming model in which data locality can be managed explicitly by the programmer. This management is achieved by explicit calls
Input file structure
In order to run NWChem calculations an input file has to be prepared, which is a free-format text file that contains start-up directives, definition of the chemical system, specification of various parameters for the calculations, and task directives. The actual processing of the input file is performed in several phases using the input parsing module. The first phase consists of processing the start-up directives that define the general features of the calculation, including available memory,
Density-functional theory calculations
Quantum-mechanical calculations of chemical systems using density-functional theory (DFT) [5], [6] is one the most broadly used capabilities of NWChem. DFT provides a good mix of efficiency and accuracy and is applicable over a wide range of chemical and material systems containing up to a thousand atoms. The efficiency of the DFT approach stems from the fact that, similar to Hartree–Fock methods, the many-electron problem is cast into the solution of single-particle equations, [5], [6], [7]
Coupled cluster calculations
The coupled cluster theory [78], [79], [80], [81] (CC) is considered by many to be a gold standard for accurate quantum-mechanical description of ground and excited states of chemical systems. This accuracy, however, comes at a significant computational cost which limits CC calculations to systems containing under hundred atoms with a total time to solution for single point calculation approaching several hours. The computational expense of CC calculations comes from the fact that it is a true
Classical force field calculations
The molecular dynamics (MD) module in NWChem provides a highly scalable parallel framework for molecular simulations based on classical potentials or force fields. The default force field is based on the Amber-type [125] representation: The first three terms describe bonded, angle, and dihedral interactions, and the electrostatic and van der Waals interactions are contained in the last two terms. The molecular
Combined quantum-mechanical molecular mechanics calculations
The combined quantum-mechanical molecular mechanics (QM/MM) approach provides a simple and effective tool to study localized molecular transformations in large scale systems such as those encountered in solution chemistry or enzyme catalysis. In this method an accurate but computationally intensive quantum-mechanical (QM) description is only used for the regions where electronic structure transformations are occurring (e.g. bond making and breaking). The rest of the system, whose chemical
Concluding remarks
New capabilities and improvements in the parallel algorithms are continually being added to NWChem to enable computational modeling of large and complex scientific problems using latest high performance hardware. Thus the development strategy for NWChem in the coming years will be focused on parallel performance, science capabilities, and open source. In the area of parallel performance the focus is on the efficient utilization of petascale and upcoming exascale architectures. This includes
Acknowledgements
The funding sources for the development of NWChem and the parallel software tools over the lifetime of the project include the division of Chemical Sciences, the Office of Basic Energy Sciences, Mathematical, Information, and Computational Sciences division of the Office of Advanced Scientific Computing Research, the Office of Naval Research, the U.S. DOE High Performance Computing and Communications Initiative, the Environmental and Molecular Sciences Laboratory, the Construction Project,
References (136)
- et al.
Physics Letters A
(1997) - et al.
Chem. Phys. Lett.
(1996) - et al.
Computational Materials Science
(2003) Chem. Phys. Lett.
(1980)- et al.
Chem. Phys. Lett.
(1996) - et al.
Chem. Phys. Lett.
(2004) - et al.
Chem. Phys. Lett.
(1995) - et al.
Chem. Phys. Lett.
(1999) - et al.
Comput. Phys. Comm.
(2002) - et al.
Comput. Phys. Comm.
(1992)
Comp. Phys. Comm.
Chem. Phys. Lett.
Chem. Phys. Lett.
Chem. Phys. Lett.
Chem. Phys. Lett.
Chem. Phys. Lett.
Chem. Phys. Lett.
Chem. Phys. Lett.
Journal of Supercomputing
International Journal of High Performance Computing Applications
J. Comp. Chem.
Phys. Rev. A
Density Functional Theory
Density-Functional Theory of Atoms and Molecules
Phys. Rev. B
Phys. Rev. Lett.
Phys. Rev. B
Phys. Rev. A
Phys. Rev. B
J. Phys. Chem.
J. Chem. Phys.
J. Chem. Phys.
Phys. Chem. Chem. Phys.
J. Phys. Chem. A
J. Phys. Chem. A
Rev. Mod. Phys.
Electronic Structure: Basic Theory and Practical Methods
Theory and Applications of Computational Chemistry: The First Forty Years
Electronic Structure Calculations for Solids and Molecules: Theory and Computational Methods
Proc. R. Soc. London, Ser. A
Journal of Chemical Information and Modeling
Modern Quantum Chemistry
J. Comp. Chem.
Introduction to Computational Chemistry
J. Chem. Phys.
J. Comp. Chem.
J. Chem. Phys.
Cited by (0)
- ☆
This paper and its associated computer program are available via the Computer Physics Communications homepage on ScienceDirect (http://www.sciencedirect.com/science/journal/00104655).