Regard to Ray’s reply, the time step is actually 0.0002 ps (0.2 fs), instead of fs, i’m sorry. I do use the same system for both qeq simulation and fixed charge simulation. The original fixed charge model has charges for Si and O 0.7 and -0.35, respectively, but the qeq charges are around 0.38 and -0.19. I think that is why the etotal differs so much in these two cases. In the original fixed charge model, the charges for Si and O (0.7, -0.35) are obtained as a fitting result of adsorption energy of small hydrocarbons in zeolites. Apparently, in the regime of equal electronegativity, the charges are far away from that fitting result.
I also tried time step 0.1 fs, the resulting etotal and temperature goes the same way as of that with time step 0.2 fs.
That is where the problem is – the potential parameters you are using are parameterized/optimized with 0.7 and -0.35 charges on Si and O. This is an important contribution since the Coulomb attraction between these two charge values balances other parts of the potential. With the charges changed to something else, the Coulomb also changes, then the entire potential energy surface changes. Therefore you are left with a set of parameters that is not optimized for SiO2, hence the energy conservation problem.
To run a test, just use the set command to change your charge values to 0.38 and -0.19 and run a fixed charge equilibration – you should see a similar trend compared to the qeq run.
What you should do now:
Fit the qeq parameters so the equilibrium charges are 0.7 and -0.35, or
Use the same set of qeq parameters, but re-fit/re-optimize the potential parameters.
Without doing either one of the above, you can’t use the potential with qeq.