Over the last few decades, modern numerical methods have reached a level of complexity where state-of-the-art algorithms can no longer be implemented from scratch in the typical lifetime of a research project or the duration of a graduate education. Indeed, implementation of a new method and comprehensive testing may exceed the time available to most projects for anything but simple model problems. Furthermore, the resulting codes are usually not or not sufficiently documented and frequently do not follow best practices in software design. Thus, it is rarely possible for researchers subsequently working on a similar problem to benefit from the previous effort.
This situation risks isolating research on mathematical algorithms from the applied sciences because the former is no longer able to demonstrate new and powerful developments using problems the latter care about. The only solution to overcome this schism is to promote reusing code. The Archive of Numerical Software addresses the current lack of venues to publish descriptions of well-documented and well-tested implementations of modern numerical methods, based on available software libraries. It highlights the creative achievement behind good implementations. It encourages the use of high level tools and reuse of existing code in building new applications. All articles published in the Archive of Numerical Software need to be accompanied by a complete set of programs including instructions how to compile and run them, such that their results can be verified by readers. They must be provided under an OSI (Open Source initiative) approved Open Source license in order to encourage their reuse as a basis for developments by others.
The Archive of Numerical Software focuses on programs built on Open Source libraries that are known to follow high implementation standards, tested software design and extensive documentation.