Fix heat pays no attention to fix shake, but it doesn't need to. I would
say the issue is simply that if you are running a non-equlibrium MD model
with fix heat, you cannot expect it to precisely conserve energy forever.
A 1 degree change in 6 psec (6000 steps) is not a lot. You might
try a smaller timestep and see if that helps. Else do some
I agree that it’s not a lot, but it is significant and it adds up over time. I tried decreasing the timestep to 0.0001 ps and the rate of temperature increase remains the same. Also, it is puzzling that a simple LJ fluid has no such problems.
I don't know. SHAKE can impose drastic forces to keep the bond
lengths requested. But the net force it imposes is 0.0. Maybe
Paul has ideas about using fix heat with shake.
I don't recall ever trying fix heat with fix shake, but it seems like it should work without a problem. One thing you might test is to see if the problem completely goes away if you simply turn shake off and let the H2O bond lengths fluctuate. 0.5 fs timesteps should be OK.
If you don't want the energy to increase, make sure you're removing just as much energy from the system as you're putting in. Even so, there's no guarantee that the T won't rise, especially if you're running NVE. Probably not recommendable to run NVT on the same group as you're imposing fix heat on, because it will likely make the fixes compete against each other.