My Community

Hello,

I am training a force field for water. As the intended temperature for production is 300 K I use a temperature ramp from 50 to 500 K to sample structures. This follows the general rule to exceed the target temperature by some 30%, but in this case I also exceed the boiling point of the material. I wonder if this is a good idea, as I see a distinct increase in the error of the energies above 400 K, the density also decreases more sharply in this temperature range, but the simulation is still far from a gas phase.

Should I stop training at the boiling point or is it fine (or even beneficial) to sample beyond this point?

Dear akretschmer,

Thank you for your question.

Indeed, it is recommended to run the training at a higher temperature, but it is unnecessary to heat water all the way up to 500 K. The error does increase with the temperature, but you can also decrease the time step and/or increase the mass of hydrogen. Considering that the production temperature is 300 K, it should be sufficient to go up to 350 K in the training.

Best wishes,
Alexey