JE0: On reproducible researchΒΆ

An article about computational science in a scientific publication is not the scholarship itself, it is merely advertising of the scholarship. The actual scholarship is the complete software development environment and the complete set of instructions which generated the figures.

—J.B. Buckheit, and D. L. Donoho

Reproducible research is an essential component of scientific discovery. Unfortunately, a lot of computational physics research is not reproducible. Very often results and figures in published papers are hard or even impossible to recreate. This makes it difficult to compare algorithms, understand and recreate published results and slows down the process of scientific discovery.

Maintaining a simulation journal, analogous to a laboratory notebook, may promote reproducibility. A simulation journal can contain more details than a published paper, provide annotated input files and code needed to produce tables, figures and results of the calculation. The journal will be most useful to the researcher who did the original simulations or close colleagues and those wanting to use the algorithms and results in their own research.

What this means in practice is that for all computations we need to store the following information.

  • For each simulation, the complete input needed to run the calculation.
  • For each simulation, the scripts/programs and data required to create the results (tables, figures, etc).
  • A journal entry describing initial conditions, boundary conditions and any special notes on running the simulations. There need not be one entry per simulation, but each simulation must be described in some entry.
  • The results presented and described in as much detail as needed.

The journal entry should written in a manner that it can be eventually incorporated into a publication and should be explicitly written with reproducibility in mind.