Fast use chemical numerics methods: the use of 'vectorization by gridpoint'

摘要

Conventional numerical techniques for integrating the gas-phase chemistry portion of a pollution-transport model tend to require a large amount of processing time to produce a model prediction. Such models often make use of a vectorizing supercomputer in order to generate output in a reasonable amount of time. However, the mathematical methods frequently fail to take advantage of the vectorization capabilities of these computers, with a resultant loss in processing efficiency. Recently, it has been shown that the time required to simulate gas-phase chemistry can be greatly reduced for the case of a highly accurate predictor-corrector method (Jacobson and Turco; Gear). The key concept employed was that of "vectorization over gridpoints", in which mathematical operations are carried out on large numbers of model gridpoints simultaneously. In this study, vectorization by gridpoints has been applied to another conventional method of chemistry integration, and comparisons are made between the original code, the improved version, the code of Jacobson and Turco, and an unaltered Gear solver. Test cases of 1195 sets of initial concentrations and reaction rates are integrated by each method, and the results are compared for both accuracy and processing efficiency. The chemical mechanism employed for the tests is that of Canada's Regional Acid Deposition and Oxidant Model. The use of vectorization by gridpoints is shown to decrease chemistry processing time by over an order of magnitude for the methods tested. For the ADOM chemistry solver, the technique removes the gas-phase chemistry's effect on the total CPU-time; the other terms in the net differential equation dominate following the introduction of the new code.