Trends in Chemistry
Volume 3, Issue 9, September 2021, Pages 697-699
Journal home page for Trends in Chemistry

Forum
Automation in DFT-based computational materials science

https://doi.org/10.1016/j.trechm.2021.07.001Get rights and content

Automation simplifies the use of computational materials science software and makes it accessible to a wide range of users. This enables high-throughput calculations and makes it easier for non-specialists to enter computational materials science. However, increasing automation also poses threats that should be considered while interacting with automated procedures.

Section snippets

Automation

Automation can be defined as the performance by a computer of a function previously performed by a human [1]. This already implies that the specific meaning of automation will change with time and development in a particular field. Automation is playing a growing role in the fields of human–computer interaction [2] and computational materials science and chemistry [3]. Today, automation in computational materials science refers mainly to the use of workflows that allow the computation of

State of the art

Many tasks in computational materials science using DFT-based quantum engines are already automated. DFT-based calculations typically require only one input file and then the calculation is performed automatically using the specifications made in the input file. Use of default values and only limited experience with DFT methods or the specific program can, however, easily lead to problems and some of these might be solved with automation [4., 5., 6.]. Other use cases for DFT-based quantum

Pros and cons

In the past, descriptions of methods and evaluations were often only incompletely documented in the corresponding publications. A full automation from the creation of the structures up to the evaluation makes the whole research process clearly more reproducible. In addition, meaningful default values for the specific task (material and overall task) can be set. It might also facilitate a switch to the most suitable package for a task and lead to not having to rely on a package that is less

How can we automatize?

Typically, additional (Python-based) programs are built for automation. They allow the building of complex computing workflows, management of queuing, and the running of jobs on supercomputers [5,6,9]. Efforts to unify certain standard tasks (e.g., structure optimizations) with different program packages have also been made [5,6]. For most of these cases, however, complex interfaces to each simulation software and for different tasks have to be written. In some cases, the developers of

Concluding remarks

Automation makes materials science on the computer much easier. We may be able to answer complex chemical questions completely automatically when the evaluation and, partially, the interpretation of the results are also automated. The solution of entire research tasks by non-specialists in the field will require much less training. Automation should be considered for the usability of new software in this area. Nevertheless, there are risks associated with this type of automation, especially

Acknowledgments

I thank Philipp Benner and Silvana Botti for helpful comments on the manuscript. Additionally, I acknowledge Gabriele Peters’ course on human–computer interactions at Fernuniversität Hagen.

Declaration of interests

No interests are declared.

References (15)

There are more references available in the full text version of this article.

Cited by (0)

View full text