Reproducibility in Science, Computer Science & Climate Science

Thomas Ludwig (DKRZ)

In the last recent years we see an increasing discussion of the topic of reproducibility on our HPC conferences. The scope is comprehensive and covers all layers of abstractions. Processor internals form the hardware layer and reproduction of scientific results at the top level is a crucial factor for any political decision making. Inbetween we find the issue of bitwise reproducibility of numerical simulation results. The talk will discuss problems in the field of computational climate science. And no, the talk will not present solutions. Instead it opens a wide field of new research questions.