Abstract
Computational techniques have revolutionized many aspects of scientific researchover the last few decades. Experimentalists use computation for data analysis, processingever bigger data sets. Theoreticians compute predictions from ever morecomplex models. However, traditional articles do not permit the publication of bigdata sets or complex models. As a consequence, these crucial pieces of information no longer enter the scientific record. Moreover, they have becomeprisoners of scientific software: many models exist only as software implementations,and the data are often stored in proprietary formats defined by the software.In this article, I argue that this emphasis on software tools over models and data is detrimental to science in the long term, and I propose a means by which this canbe reversed.
Highlights
Computers have become an essential tool in many aspects of science: they help with collecting and processing data from observations, evaluating theoretical models, and communicating with fellow scientists
The situation I have described is a symptom of a lack of exchange between the natural sciences and research in computer science
Today’s computational scientists see computer science as an engineering discipline that provides them with ever increasing number crunching power. Their own training in computational techniques is usually limited to managing the practicalities of working with software tools
Summary
Computers have become an essential tool in many aspects of science: they help with collecting and processing data from observations, evaluating theoretical models, and communicating with fellow scientists. Software as a notation for scientific knowledge In the previous section, I have explained the undesirable consequences of the fact that computational models are often inseparably intertwined with the software tools that work on them There is another important problem resulting from the fusion of tools and models, which is related to the different time scales on which science and computing technology evolve at the moment. A scientific model specified with precise floating-point semantics cannot be implemented correctly using today’s scientific programming languages This situation is a consequence of the attitude that I have described in the introduction: computational science is so much focused on the performance of the computations and so little on the correctness of the results that there is no incentive for language designers and implementors to improve the situation. Suitable domain-specific model languages and tools that work on them remain to be developed
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call