Why is the planewave codes ecosystem healthier on free/opensource software adoption, when compared to codes supporting atom-centered orbitals?

Question

Regarding the branch of matter modeling that deals with electronic structure methods, I think we can divide it in two camps: Those who work mostly with planewave-based software, geared towards periodic systems, typical of solid-state physics research; and those working mostly with software based on atom-centered Gaussian type orbitals, geared toward molecular systems, typical of quantum chemistry research. There's a lot of overlap between both camps, but I think people tend to stick mostly with one kind of system, in their research.

Said that, as someone more familiar with systems typical of quantum chemistry, even before having the hard numbers to back it up, I always had the impression my peers at the solid-state physics community were better served in terms of free software / open source tools (and even just "free as in beer" tools), while we were mostly stuck to expensive proprietary tools, the alternatives being much less well known and used. I even felt a bit envious because of this.

Now I noticed this answer to one previous question here: "More matter modeling codes or better codes?", where user @leopold.talirz pointed to a study preprint(1) where they systematically reviewed available matter modeling packages, their citation counts, source availability (either closed, available, copyleft or permissive) and cost (either commercial, free, or free only for academics). Then I actually downloaded the data in their site, to check if my impression about this was right or not. I think it is mostly correct. If we filter by the "PW" tag, for planewave, we get this:

While filtering by "GTO" tag, we get this scenario:

As we see, while both graphs show commercial software with high counts, the one for software with PW tags shows a sizeable fraction of total citation counts associated to noncommercial free / open source codes, specially to the gold standard copyleft licenses. Note the absence of opaque closed source codes. So the software ecosystem here looks healthier. For software with GTO tags, the share of free / open source is much lower.

So I ask, what are the causes for this matter modeling software freedom gap between PW-based codes and GTO-based codes?

References:

(1). Talirz, Leopold, et al. “Trends in atomistic simulation software usage”. arXiv:2108.12350 [cond-mat, physics:physics], agosto de 2021. arXiv.org, http://arxiv.org/abs/2108.12350.

Answer 1

Disclaimer: These are just some observations; I don't claim to know the answer to this question.

There's a lot of overlap between both camps, but I think people tend to stick mostly with one kind of system, in their research.

You may be largely correct - I'd just like to point out that if we are looking at the simulation engines (rather than what people do) there now are some that marry both worlds, such as CP2K or NWChem.

what are the causes for this matter modeling software freedom gap between PW-based codes and GTO-based codes?

Some possible factors to consider:

• The citations to commercial codes in this dataset are strongly dominated by Gaussian on the GTO side and VASP on the PW side (together they account for >50% of citations to commercial codes). Findings may therefore be difficult to generalize.

• Quantum chemistry programs have had a substantial head start on solid state. The first version of Gaussian was released 1970; for VASP I don't know the date of the first public release but I guess at some point in the late 1990s depending on the definition of the term.

• On this time scale, open-source in scientific software is a relatively recent development - the GPL licenses were published in the 1990s [1].

[1] While the MIT license was created earlier (in the 1980s), the popularity of permissive licenses like MIT has only very recently (late 2000s) gained more prominence, at least as far as the codes in this dataset are concerned.