Water is an interacting society, not a molecule, and it is a society of related molecules, not just H2O; among those molecules H2O is just the most prominent family, not a single kind of molecule; and light water (what we usually think of as H2O) is just the most prominent branch of that family.
So I was interested to see a recent science news item come up about studies of the molecular structure of water:
A study published in Nature Physics provides new molecular-level evidence from simulations that liquid water is not a single uniform substance, but a constantly shifting mixture of two distinct microscopic structures.
As the article notes, this is not a new idea. It's not surprising, because it confirms what chemists and physicists had already suspected. For reasons tied to the behavior of water, it has long been hypothesized that water molecules shift around between two kinds of structures. This is known as the two-state model, and the reason why it was already taken to be probably true is that water seems schizophrenic, sometimes acting as if it were highly structured and sometimes as if it were very loose and disordered, but rarely in between. Naturally the easiest explanation of this is that it actually sometimes is highly structures and sometimes quite disordered, and rarely in between; and what we know of water molecules doesn't rule this out. The big puzzle was that it was very difficult to find any evidence of this, beyond the behavior that led to the model in the first place. One might think that you could confirm it by just checking how the molecules are related to each other, but it's actually quite difficult to get good direct evidence of the overall structures of vast numbers of molecules in constantly changing relationships. And in particular there was no clear idea of how the two states would even be related to each other in the actual behavior of water.
So, as I understand it, having looked at a number of these pop science articles and having briefly skimmed through what I could find of the relevant papers without forking any money over to get behind paywalls (and without pirating anything!), the state of things is something like the following.
The new discovery is repeatedly called 'direct molecular evidence for the two-state model', in this and other articles, and it is, but in a more complicated than I think that makes it sound. Nobody has seen these two structures themselves, and there's a lot that is still unknown. The two-state model, to work at all, requires that there be a state of water known as liquid-to-liquid transition, in which water splits into a high-density state and a low-density state. It wasn't obvious that water had such a transition because you can only get it under certain conditions, in a very, very, very cold state in which liquid water is still liquid but is almost right at the edge of becoming solid. This is very difficult to set up, because water freezes so easily, and usually does so long before this point, so you have to supercool the water in such a way that it stays liquid. Computer simulations suggested that something like this state probably existed, but even the best computer simulations were quite messy; physical chemists could guess where it would have to exist if it did exist, but they had to make assumptions that were not yet proven; and nobody had yet been able to produce it in a way that got useful measurements. Improvements in computer simulation, however, made it possible to give a more precise statement of the relevant numbers for measurements related to energy and density under certain conditions, and the answer was coherent, and also consistent both with the two-state model and what else we know of water. And, without trying to make the model fit some more recent evidence based on new technology and measurement processes using X-ray lasers (which we began to get earlier this year; see here also ), the model happens to fit new measurements that we were only just recently able to obtain.
Thus the 'direct molecular evidence' is that the two-state model now has clear and definite computer simulation of water that shows us how, given everything else we know about water, the two states would have to be related to each other, and new measurements under roughly the conditions expected by the model are consistent with what the precisified model says we should get. That is, the two-state model went from being a rather hazy idea with a lot of guesswork to being a fairly precise idea with definite features, and, importantly, when the hazy idea was made precise and definite, it fit both the old evidence and some new measurements we were just now able to start getting. William Whewell would really like this example, because he was always a champion of the idea that one of the routes of progress in the sciences was just making ideas more clear and precise; here, the computer simulation is evidence of the two-state model because it makes the two-state model much more clear and precise without breaking it in any way, despite measurements that have been improving at the same time. Nobody has seen or traced out the actual structures (because it is still very difficult to do); the evidence is not direct in that way. We just have new direct measurements that confirm their existence, in exactly the way the theory indicates that they should.
What is perhaps almost as important is that a new line of inquiry was opened up by this. The computer simulation indicates that the two structures change their behavior depending on the temperature and pressure, in ways that we might be able to confirm or disconfirm in the future.
Thus the two-state model, our best idea of how water might work, is almost certainly right, although there is still room for unexpected discovery. Thus water is not only an interactive society of many different molecules, of which H2O is only the most important, but an interactive society whose molecules are related in at least two different ways, each of which contributes to different aspects of water's behavior. Water is a surprisingly deep subject.