Atomistic simulations are widespread in modern chemistry and materials research as complements to experiments. They are fundamental to interpreting and giving new insight into experimental data, predicting new materials, and screening for desirable properties. These methods encompass a range of modeling approaches from workhorse classical interatomic potentials and density-functional theory (DFT) to reference-quality post-Hartree-Fock methods and emerging machine-learning potentials. They are implemented across an intimidating assortment of software packages which make different trade-offs in user-friendliness, interoperability, and ease of automation. It can be difficult to trace the origin of a calculation or ensure that the results used in a study are consistent, and it can be costly to move from one research group to another with an incompatible in-house toolchain. Without shared collaborative tools, high-level algorithms may be implemented in a particular package, leaving the rest of the community to create in-house variants for their own work.

The Atomic Simulation Environment (ASE) is a community-driven Python package that solves the "n² problem" of code interfaces by providing some standard data structures and interfaces to ~100 file formats, acting as useful "glue" for work with multiple packages.^[1] ASE integrates with more than 30 atomistic codes, covering methods from classical MD, machine learning interatomic potential to ab-initio codes. This coverage of different packages it is unparalleled in our communities. In addition, it can manipulate structures and run calculations, providing a range of generic dynamics and geometry-optimisation routines with a toolkit for the development of such methods. New schemes such as preconditioned optimisers are written once and immediately available to users of established atomistic codes. ASE is used for input preparation such as generation of slabs and nanoparticles, and output analysis such as thermochemistry, phase-diagram generation and energy level plots. These features are generally omitted from academic calculation packages but benefit from inclusion in programmatic workflows. Developers of new calculation packages such as QUIP have been able to focus on novel aspects (e.g. implementation of machine-learning potentials) and make use of existing tools for structure manipulation and dynamics.

Workflow management tools are emerging to support automated high-throughput calculations and ASE is an integral part of some of them. The Atomate suite focuses on a smaller set of calculators and is used for the high-throughput Materials Project; Aiida has a plug-in system and strong focus on data provenance. The Atomic Simulation Recipes (ASR) framework is linked more closely to the ASE API, providing some data provenance/management features and using the MyQueue package for job scheduling.^[2]

The Python programming language is now taught as part of many science/engineering undergraduate courses. The emerging "data science" field has heavily used and contributed to a scientific Python ecosystem with high-quality packages for linear algebra, graphics and machine learning. We would encourage early-career researchers to engage with these libraries and use them to rapidly develop their own workflows. Funders including UKRI increasingly demand that data and software are made available alongside publications. In this workshop we hope to highlight and teach good practices around presentable, sharable, automatable calculations and analysis, using available and emerging Python-based tools. Software can be used to create very complex analysis pipelines, but if handled properly can also provide exact documentation of those methods.

applications will open soon via a dedicated website.