Publishing Reproducible Numerics: A Student's Perspective

I gave a presentation at the RDM Workshop from the SFB 1481: Sparsity and Singular Structures at the RWTH Aachen about reproducible numerics. You can download the slides here. This post is a short summary.

Reproducibility has always been a fundamental aspect of scientific inquiry. But in recent years, the inability to reproduce a substantial number of scientific findings has cast doubt on the research process. This post delves into the motivation, definitions, tools, best practices, and concludes with the need for reproducibility in research.


Status Quo: A significant number of researchers utilize software for their research studies.

Relevant Situations:

However, there's a Problem: Reproducing research is quite challenging in reality.

Nature Survey

The crux? We require strategies to enhance reproducibility. But remember, no labs are involved, only computers.

What Does "Reproducible Numerics" Mean?

Definition: It refers to the ability to reproduce the exact numerical results using an identical setup.

Why is it crucial?

But, What can go wrong?

Numerical Experiments Numerical Experiments Numerical Experiments

Tools & Best Practices

Reproducable Progress (with Git)

Version Control is achievable with Git.

Git Commit

Reproducable Environments (with Docker)

Different individuals have different setups. One solution is Virtualization with Docker.

Dockerfile Example

Reproducable Workflows (with Gitlab/-hub)

Explain how your code or tool works with a file.

Documentation Example CI Example

Archiving Code for Publication (with Zenodo)

When your code and paper are ready for publication, DOI (Digital Object Identifier) System comes to the rescue.



In the end, for reproducible:

The importance of reproducibility cannot be overstated in research. With the tools and best practices discussed, ensuring reproducibility becomes an achievable task.