Home » Technology » Scientific Computing » Chemistry in the Fast Lane

Chemistry in the Fast Lane

© Adam Hughes

Chemistry in the Fast Lane ------------------------------------------------------------- As discussed in a previous article, computational chemistry can be an extremely powerful research tool, either as a stand -alone method or in conjunction with experimental procedures. Because of the accuracy demanded in this field, however, the computational details involved in a chemical simulation can become very intricate and correspondingly costly in terms of computer time required. To overcome these time limitations, scientists have developed numerous approximations and "tricks" which make a problem more manageable yet maintain the overall reliability of the calculation. As successful as these efforts have been, however, they often involve a good deal of research in themselves and can be slow in developing. In order to achieve truly significant gains in calculation speed (orders of magnitude!), researchers in computational chemistry are focusing their efforts in earnest on integrating new computational methods with with the powerful computing resources widely available today. The most striking example of this interplay of code engineering and state-of-the-art hardware is in the application of parallel computing to computational chemistry.

Probably the most straightforward example of parallel computing in chemistry is the spatial decomposition technique in a classical simulation. In a serial classical simulation, the chemical system is represented by a set of coordinates, and the processor must calculate the forces on each particle at every time step. Obviously, as the system size grows, this process can become quite laborious and time-consuming. A spatial decomposition essentially divides this work among a given number of processors, thus reducing the compute time for each. Using parsing algorithms, this method assigns a certain portion of the sample space to each processor. The processor then performs all calculations necessary, but only for the atoms contained in its defined volume. From time-to-time, depending on the criteria dictated by the problem, processors may need to communicate with their immediate neighbors to exchange updated information about particles which interact, but lie in different (but neighboring) regions of space.

In the absence of any communication steps, the example parallel scheme discussed above could be expected to yield nearly perfect parallel speed up, modulus any disk/swap memory accessing issues. That is, if processors didn't need to talk, a job run on eight processors would finish about eight times faster than the same simulation run on one processor. However, communication is a real and limiting factor in these types of simulations, and it is often necessary to optimize the number of processors used for each problem to be run. Generally, there is some "sweet" number of particles per processor, where there is a substantial number of calculations

The copyright of the article Chemistry in the Fast Lane in Scientific Computing is owned by . Permission to republish Chemistry in the Fast Lane in print or online must be granted by the author in writing.