I am currently preparing to study my material using the CVFF force field, but I found that some bond and angle parameters (such as ether bond/angle) are missing in the force field. Therefore, I obtained the relationship between energy and bond length/angle degree through DFT calculations and then calculated the bond and angle coefficients by fitting them with a harmonic potential.
I found that the angle coefficient is very small, only 0.02 kJ/mol, much lower than the typical angle coefficients in the CVFF force field, such as 90 kJ/mol. In this case, I can’t use a 1 fs time step for calculations and have to reduce the time step to 0.1 fs. Even so, the system still encounters errors, such as bond/angle atoms missing, after running stably for a period (10 ns).
Does anyone have any suggestions to keep the system running stably? Even a 0.1 fs time step is acceptable for me. I am currently considering increasing the angle coefficients by 10 times for testing. I tested by amplifying these parameters 1000 times, bringing them to the same order of magnitude as those in CVFF. Although amplifying the parameters allows the system to run stably with a 1 fs time step, it leads to an inability to accurately predict the material’s melting behavior, viscosity, and wettability.
Rather than making up your own protocol, you should study the publication(s) for the CVFF parameters and follow the “recipe” that is given there to produce the CVFF parameters.
That should also give you suggestions for how to validate your parameterization.
Are you sure there is no mistake in converting units?
Why first running DFT calculations when you later apply some arbitrary scaling factor. Then you could scrap the whole process and substitute the missing parameters with the closest fit from the available ones. That is far less arbitrary.
I doubt that a few missing bonded parameters have such a grave impact and particularly on those properties. Those are far more (yet not exclusively) determined by partial charges and non-bonded parameters.
Correctly determining the melting point from MD simulations is a particularly difficult problem since melting and freezing are activated processes and thus on the atomic scale often associated with a large hysteresis and thus require special procedures, like coexistence simulations.
Thank you very much for your reply which has given me a lot of inspiration。
Yes, this is exactly what I need to do. I am already gathering original literature on the construction of the CVFF force field and hope to gain some guidance from it.
I checked the data processing procedure. I did pay attention to the unit conversion issue, and the energy unit was correctly converted from Hartree to kcal/mol. I am now quite confused and wondering if there is an issue with my DFT calculations. I have attached the results of the flexible scan of the angle for the ether. If you are willing, could you help me check it? Additionally, in Prof. Adri C.T. van Duin’s paper comparing ReaxFF and DFT calculation results (DOI: 10.1021/acs.jpca.6b12429), I also observed similarly low angle energies. However, I did not find angle coefficients as small as those I obtained in my calculations. 4-12-13.fchk (5.3 MB)
Yes, you are right. If we arbitrarily change the parameters obtained from the calculations, then the calculations become meaningless.
To be honest, due to the excessively small parameters obtained from the DFT calculations, I was unable to complete the wetting simulation. When I used the amplified parameters to perform the wetting simulation at a temperature where good spreading was observed in experiments, the droplet did not spread at all. Additionally, I calculated the viscosity using the DFT parameters, which was very close to the value measured in my experiments. However, when I amplified the parameters, I could no longer obtain results close to the experimental values. Therefore, I concluded that bond angle parameters have a significant impact on viscosity and wetting. I think my viewpoint was too arbitrary. I should take some time to think carefully about how to solve the current issue.
Finally, thank you once again for your very valuable response, and thank you for your consistent contributions to the community.
