B26. Convergence-based Solution Correctness Testing

Poster Title

Convergence-based Solution Correctness Testing

First Author

Shixuan Zhang

Topicatmospheric model development, computing, E3SM evaluation, software tools
AffiliationSciDAC
Link to document


Title

Convergence-based Solution Correctness Testing

Authors

Shixuan Zhang, Hui Wan, Phil Rasch, Balwinder Singh, Kai Zhang, Vince Larson, Carol S. Woodward

Abstract

E3SM solves a large set of equations in millions of grid cells. A typical comparison of two solutions involves comparing “climate statistics” of tens to hundreds of physical quantities. Multiple years of simulations are needed to produce data to derive such statistics, costing a lot of computer time. Simpler, cheaper, and more objective methods are needed for the testing of solution correctness, especially for high-resolution simulations. We have developed a simple, quantitative and objective testing method using the concept of time step convergence and used it to assess solution correctness in new software and hardware environments and for simulations using reduced floating-point precision. Results from short convergence tests are found to be good predictors of E3SM’s long-term behavior related to fast physics, making the test computationally very efficient. The method is expected to be applicable to other models that solve time evolution equations.