One of the issues is the results from molecular simulations depend on the interatomic interaction model as well. For solubility of gases, good chances are the results will be all over the place as well unless the models for water and gases are tailored specifically to reproduce the experimental data.
For example, see the comparison between TraPPE and OPLS force fields in Fig. 9 here.
1 Like
Just out of curiosity - how well liquid water can be currently simulated, in general, and by Molly.jl specifically? I mean, water is both very well experimentally studied, and pretty complex liquid: with hydrogen bonds, dissociation etc. And - if some reasonable (i.e. fitting ALL reliable experimental data that are available) simulation is possible, what are the computation resources necessary?
Non-polarisable, 3-point, molecular mechanics water models are cheap but struggle to match all properties. 4-point and other virtual site water models can match more properties. Here are some relevant papers:
Generally the properties matched are in the conventional temperature and pressure ranges. Water models that match for example all known ice structures are beyond my expertise, but if they exist they are expensive.
All 3-point water models are compatible with Molly. We don’t have support for virtual sites, yet, which means we don’t support 4-point water models. We also don’t support polarisable models like AMOEBA yet.
Out of curiosity, when simulating water – is it realistic to use Lennard-Jones potential?
I looked into McQuarrie and Simon’s 1999 book, and found some values for \sigma and \varepsilon. These data do not include \sigma and \varepsilon for water.
They also discuss the origins of attractive forces, with potentials for London dispersion, dipole-dipole moment and induced dipole moment. From these, it is possible to find \varepsilon\cdot \sigma^6. I have computed estimates of \varepsilon when (i) assuming that \sigma is known, and (ii) when I replace \sigma with the collision diameter.
It seems like \hat{\varepsilon} computed from London dispersion, etc. is decent for some substances, but deviates considerably from the listed values of \varepsilon for other substances.
Does it make sense to use my estimates of \sigma \approx d_\mathrm{coll} (collision diameter) and \hat{\varepsilon} for water??
Out of curiosity, when simulating water – is it realistic to use Lennard-Jones potential?
No, this wouldn’t lead to satisfying results.
For modeling the solubility of relatively common systems like water-O2, you should get good results with EOS models (or similar) using Clapeyron.jl. Using molecular simulations here would be overkill (without getting better results). How did you calculate the solubility in Clapeyron (which model and method)?
2 Likes
You might find this paper relevant, 10.1016/S0304-386X(98)00007-3
But where is water there?
I mean, in the “classical” case it’s a pretty concentrated alkaline solution, and that’s not water.
In the “modern” case: the water is in the polymer membrane, and maybe also as a surface layer. That’s also not water.
1 Like
As mentioned previously, I used the same EoS for both phases. I tried with Peng-Robinson, some empiric Helmholtz EoS:s (MultiFluid, GERG2008), as well as various SAFT models. The give good fit to experimental data for the vapor phase, but relatively poor fit for mole fraction O2, etc. in the liquid phase. A miss of a factor of 30-50 or so, I would guess.
I assume this is because these models are not “trained” in particular for the liquid phase, while a Henry constant model is. Perhaps it is possible to get better result if I use different models for the liquid and vapor phases?
There is lots of water and oxygen in the anode separation tank in an electrolyzer. I’m not thinking of in the membrane.
1 Like