Coadsorption in framework

Hi Máté,

Have you looked at the fluctuations in the number of molecules in your simulations? In the files you uploaded, the number of TCP molecules are not changing. That suggests your loading results are highly correlated with your initial conditions, not that the ordering of the species is influencing your results.

Ryan

Hi Máté,

"am doing something wrong in the inputs?" Your input files look fine.

"I would have expected them to converge to the same regardless of the different seeds (statistically they should, right?)" Yes, equilibrium loading is independent of random number seeds, ordering of species, and initial configuration. However, your initial configuration can effect how long it takes your simulation to reach equilibrium. From your files, it appears that your initial configuration (an empty framework) is quite far from equilibrium. In contrast, you could choose to start from any of the following initial configurations:

• a framework filled with 1500 water molecules
• a framework filled with 250 TCP molecules
• a framework filled with 1000 water and 100 TCP molecules

All should relax to the same equilibrium loading, but some may take longer than others to get there. The relaxation time may even differ from run to run, as you are seeing with your sims. It's generally good practice to start multiple sims from different compositions and run until they converge to the same equilibrium composition.

Looking at the property files you uploaded, the framework rapidly fills with molecules in the first 10 million MC steps. However, since Cassandra chooses the order of attempted insertions randomly, one of your sims fills first with water molecules, leaving relatively little room for TCP molecules. The other sim fills first with TCP molecules, and as a result has much less room for water. Plotting the number of TCP molecules vs MC step shows the fluctuations in TCP molecules drops drastically after 10-20 million steps. From that point onward, your sampling of the composition of the adsorbed fluid appears to be very slow.

Ryan

add_to_config will allow you to cram way more molecules into a system than are thermodynamically reasonable. I tried using it to add 50 TCP molecules to the faujasite framework, and it slotted three TCP molecules into a single zeolite cage. As you can imagine, that's a high energy configuration, but due to the pore structure of faujasite even a short NVT ensemble couldn't shake those 3 molecules apart. Sure enough, when I started a GCMC sim, it crashed when attempting to delete one of those 3 TCP molecules. The problem lies in trying to compute the reverse probability for inserting that molecule back into the location it's being deleted from. If the reverse move is more than 708kT uphill, then the acceptance probability of the move blows up.

If you really want a faujasite crystal loaded with TCP to start your sim, I think you'd be better off running a short GCMC simulation without water and with a high chemical potential for TCP. The beta*mu*N term will facilitate high loading, but the beta*deltaU term will prevent high energy configs from being accepted.

I also recommend that you run in equilibration mode so your maximum translation and rotation deltas can be adjusted. The values in your input file (5 Angstroms, 90 degrees) give relatively low acceptance (1%) of your translation and rotation moves for TCP. Also, computing the pressure in a simulation box with a rigid zeolite framework is physically meaningless and computationally expensive.