Improving the quality of scientific computing via software engineering methodologies

拟议研究计划的目标是提高科学计算(SC)软件的质量。虽然已经实现了许多高质量的算法，特别是在正确性、可靠性和性能方面，但还有其他一些特性，如可用性、可维护性、可验证性、生产率、可重用性和可移植性，往往被忽视。为了解决这些质量问题，建议的方法是采用已成功应用于其他类型软件的软件工程(SE)方法。可以从SE改编的一个新想法是将SC码作为一个程序族来发展。在这种方法中，类似的软件产品被一起开发，公共部分被重用，可变功能被识别和系统地处理。这项研究有可能显著提高SC码的可重用性，因为公共码只需实现一次。测试是SE实践的另一个领域，将根据提高SC软件质量的目标进行调整。测试研究将是具有挑战性的，因为大多数测试方法假设每组测试输入都可以确定预期的输出，但SC软件通常不是这样。这一挑战将通过采用独立算法的并行测试、与封闭形式解的比较、使用区间和无限精度算术、后验误差估计以及系统地识别和记录解的已知特征的技术来解决。

The objective of the proposed research program is to improve the quality of scientific computation (SC) software. Although many high quality algorithms have been implemented, especially with respect to the qualities of correctness, reliability and performance, there are other qualities, such as usability, maintainability, verifiability, productivity, reusability and portability that are often neglected. To address these quality concerns, the proposed approach is to adapt software engineering (SE) methodologies that have been successfully applied for other types of software. One new idea that can be adapted from SE is the development of SC code as a program family. In this approach similar software products are developed together, with the common parts reused and the variable features identified and systematically handled. This research has the potential to significantly improve SC code with respect to reusability because the common code only has to be implemented once. Testing is another area of SE practise that will be adapted with the goal of improving the quality of SC software. The testing research will be challenging because most testing methodologies assume that for each set of test inputs an expected output can be identified, but this is often not the case for SC software. This challenge will be addressed through techniques of parallel testing with independent algorithms, comparison to closed-form solutions, the use of interval and infinite precision arithmetic, a posteriori error estimation and systematic identification and documentation of known characteristics of the solution.