Simulated Tempering

[Kyle Kihn]

Hi all,
I have started to try and use the simulated tempering aMD on a system. The system and context were created and then this call was used to run the simulated tempering:
st = SimulatedTempering(simulation, numTemperatures=20, minTemperature=310*unit.kelvin, maxTemperature=400*unit.kelvin)
st.step(25000000)
Upon analysis of the State Data Reporter, it appears that there are a large number of steps that drop below our minimum temperature. Is this normal or are we calling the temperature range wrong? Also how do we get it to output the calculated weights?
Thank!

[Peter Eastman]

There are two different temperatures involved. First, there's the temperature specified by the integrator. This is the temperature of the heat bath the system is coupled to. This is what minTemperature and maxTemperature control. Second, there's the instantaneous temperature of the system, which is just a measure of the kinetic energy of the system. This is what StateDataReporter outputs. The instantaneous temperature fluctuates around the temperature of the heat bath, so it will sometimes be lower and sometimes higher.

[Kyle Kihn]

Thank you for the help. One follow up question is that when we visualize the simulation runs using VMD, we see the protein (at steps the correspond with low state data reporter temperatures ~100K) basically freeze for several frames. After a while in this state it proceeds to increase in motion again. It will repeat this through out the simulation. Is this expected behavior?
Thanks again for all the help!

[Peter Eastman]

When you say that it freezes, do you mean that it truly doesn't change at all from one frame to the next, or just that it changes very little? There will of course be much less motion at low temperatures than at high temperatures.

[Kyle Kihn]

It still moves, just very little. I think what we are more concerned about is that in the frames that correspond to the little motion the temperature reported by the state DCD reporter is in the 100K range (well below our limit of 310K). We know that this is the instantaneous temperature and not the integrator temperature, but we want to make sure that a temperature this low is normal with an integrator temperature_min at 300K. We see this large discrepancy between the two temperatures a significant amount of times in our run. We are just hoping to check that this is expected and not an error on the part of our run.
Thanks again!

[Peter Eastman]

How large is your system? How many atoms and how many constraints?

[Kyle Kihn]

Our system is a single chain of a dimer protein that is solvated in a water box. The system has 70,212 atoms and the lengths of all bonds involving hydrogen are fixed. Also, rigidwater is set to true.
Thanks!

[Peter Eastman]

For that large a system the temperature fluctuations should be small, only a few degrees. Can you post your code? I'd like to look at it and see how you're setting things up. Also, on the steps when the StateDataReporter says the temperature is low, what does the output of the SimulatedTempering say the temperature is set to?

[Kyle Kihn]

Attached are the two outputs and the script we are currently using. This is from a run that is still ongoing (with the same system), as we had some issue with the output in the initial run.

simulated_tempering .txt

(14.56 KiB) Downloaded 47 times

results.txt

(116.54 KiB) Downloaded 45 times

output.txt

(320.45 KiB) Downloaded 44 times

[Kyle Kihn]

I also tried to attach the final outputs from this system as the run finished up, but the file was too large to attach.
Thanks again for all the help,
Kyle