Water Liquid-Vapor Pressure vs Density simulated by GCMC

Hi Kecheng,

It appears that at the conditions you highlighted green, your simulations are not long enough to converge to the thermodynamically stable density. To review what we should expect from a GCMC simulation:

(*) there is a value of the chemical potential, mu.equil, for which a liquid phase will be in equilibrium with a vapor phase
(*) for mu > mu.equil, the stable phase will be a liquid
(*) for mu < mu.equil, the stable phase will be a vapor

However, the length of a simulation required to converge, depends on the initial configuration. For simulations started from the vapor phase and:

(*) with mu < mu.equil, the simulation will converge quickly. These are the cases you highlighted yellow.
(*) with mu >> mu.equil, the driving force delta.mu = mu - mu.equil is large enough that the simulation length needed to converge isn't too long. These are the cases you highlighted blue.
(*) as mu ---> mu.equil from above, the simulations need to be increasingly longer in order to converge. These are the cases you highlighted green.

For this last case, I'd recommend starting from a relaxed liquid configuration rather than using make_config to generate a vapor configuration. Also, here's a few points to keep in mind:

(*) for a rigid water model, there's no need to use regrowth moves
(*) for a GCMC simulation, outputting the volume is pointless
(*) for a liquid simulation, the pressures fluctuates wildly (+/- 1000 bar) so you will need long simulations if you want to converge <P> to ~1 bar

Hello!

Could you please confirm which water model reference results are you using from NIST? I'm familiar with the reference simulation website ( www.nist.gov/programs-projects/nist-stan...e-simulation-website ) which I believe doesn't include SPC but includes SPC/E. I think the input files you sent are for the SPC water model.

Thanks!

Hi Kecheng,

"the pressure fluctuation seems very large, the total MC step reaches 1e08, is it still too short to reach euqilibrium?"

Pressure fluctuations in a liquid are going to be large, and they won't decrease with a longer simulation. There's nothing you can do to change that. Fluctuations in the average pressure, however, will decrease with simulation length (i.e., as you average more data points, the average will converge to a constant).

I'd think that the energy U, number of molecules N (or density), and U - mu * N would all be better quantities to indicate whether your simulation has converged. Can you also share what your insertion and deletion acceptance ratios are? I'd think you'd want to see several thousand insertions and several thousand deletions to indicate that you're getting good sampling.

"would 'production' be better to conduct this simulation?"

The only difference between equilibration and production Run_Types is that, in equilibration, Cassandra will change the maximum distance and maximum angle used for translation and rotation moves until the acceptance ratio is 50%. From the input file you uploaded earlier, it looks like these values start at 2.0 Angstroms and 180 degrees. In my experience, these are too large for liquid water. Cassandra will change these values every 600 moves (based on your input file) until they reach something like 1.0 Angstrom and 20 degrees (=0.35 radians). Each change will be written to the log file.

In a production run, the max_width and max_angle will not change. They will stay at whatever value is in your input file, or if you start from a checkpoint file, to whatever values are in your checkpoint file.

Ryan

Also, from your plots it appears that

(*) at mu = -45.5 kJ/mol, <P> ~= 0-100 bar
(*) at mu = -47.5 kJ/mol, <P> ~= -1100 bar

The negative pressure indicates that the liquid is under tension (i.e., the system wants to be a vapor at this mu). This is consistent with your previous result that <P> = 0.02 bar at mu = -47.5 kJ/mol. Similar to the problems you had converging to a liquid when you started from a vapor for -45.5 < mu < -40 kJ/mol, you will have slow convergence to the vapor state if you start from a liquid for mu < -45.5 kJ/mol.

Ryan