[GlibMonkeyDeath]

This is a molecular dynamics (with a specific TIP4 water model) and AI. I would characterize this paper (and basically most MD/AI simulation) as "guided hypothesis generation". I am skeptical that this result will hold up to experimental validation. It is incredibly easy to generate molecular dynamics results that look perfectly reasonable but have nothing to do with "reality" (i.e., how a lab experiment with the same system would turn out.)

One of the biggest challenges with water is modeling proton transport (i.e. pH, as water is a weird self-ionizing material.) Protons move too fast for MD steps to be stable, so they have to be approximated in some way. That is one reason why there are so many water models to choose from. Each model has a trade-off that is fine in some contexts, disastrous in others.

My money is on a TIP4 artifact.

[mefistofeles]

There's plenty of experimental evidence for similar things. That's why I made a comment on this earlier about how it relates to it. Check my comment or https://www.science.org/doi/10.1126/science.aec0018 or https://www.nature.com/articles/ncomms3401