Thanks for the responses Steve and Axel.
My understanding was that the Nose-Hoover equations of motion conserved linear momentum, and so it is not prone to the flying ice cube effect, though maybe there is something I’m misunderstanding. I’m not sure about how the charge dynamics change the equations of motion, but even if linear momentum wasn’t strictly conserved for these equations of motion, I figured that the coupling between the atom and charge degrees of freedom would be extremely small and the buildup of total momentum would be slow, but this could be a naïve view.
One thing I did forget to mention last time was that if we use fix qeq/comb and run 1 to initialize the charges and then remove the fix, the problem also goes away. We see very good energy and momentum conservation, so something about the charge dynamics is causing it.
Steve, it does appear that a Langevin thermostat does indeed remove this problem. This may be how we fix the problem. We were just very surprised by the fact that just by including the charge dynamics caused this rapid buildup of linear momentum and wanted to get a handle on its origin before we decided to switch thermostats without understanding the cause.
Thanks again for all of the help.