Question about Custom Thermostat and Barostat

[Kailong Mao]

No problem! Please see attached for the code. Thank you very much for taking the time to help us!!

Also, what I have right now is a Verlet integrator. But eventually we'd like to implement a multiple time step integrator. I know I can access a Force object and change its group by

Code: Select all

forceIndex = 3
force = system.getForce(forceIndex)

The problem is that for Force object, it doesn't have any attribute like "name". So, I don't know its identity, e.g. whether it's harmonic bond force or electrostatic. How would I know what type of force it is?

Thank you very much!

[Peter Eastman]

Thanks! I'll take a look.

You can use the isinstance() function to check the type of an object. For example,

Code: Select all

if isinstance(force, NonbondedForce):
    # Do something

Peter

[Peter Eastman]

I merged the fix:

https://github.com/pandegroup/openmm/pull/1384

Try pulling the latest code from github and see if it works correctly for you now.

Peter

[Kailong Mao]

Thanks a lot for the quick fix! I re-compiled the code and found out that although the variable "temp" now gets updated correctly, another user-defined variable "c" now doesn't. Although I added the step "addComputeGlobal("c", "exp(-dt/tautemp)")", "c" keeps its default value throughout the entire simulation. Could it be because "dt" is not a user-defined variable? The code should be in the file attached to my previous post. I was able to get around this because I realized that "c" should actually be a constant, so no need for the "addComputeGlobal" step. So, the thermostat works for me now, but I thought I should probably let you know about this.

Also, I wonder why the temperature I calculated using the following formula doesn't agree with the temperature from OpenMM.

Code: Select all

df = nAtoms * 3.0 - 6.0
gasconst = 8.314459848e-3
...
self.integrator.addComputeSum("ke", "m*v*v/2.0")
self.integrator.addComputeGlobal("temp", "2.0*ke/(df *gasconst)")

Thank you very much again for your time!!

[Jason Swails]

Could it be because "dt" is not a user-defined variable?

"dt" is the time-step and is typically constant, I think (except for some kinds of variable integrators perhaps). Were you defining your own "dt" variable for something else? Did this not result in an error?

[Kailong Mao]

Hi Jason:

Sorry for the confusion! No, I did not define my own "dt". Before Peter posted the fix, I was not able to update one variable "temp" that I defined when using "CUDA" as my platform. Everything works when I use the reference platform. Now with the fix, I can update "temp", but not another user-defined variable "c". I was able to get around this because c actually should be a constant.

Thanks!

[Peter Eastman]

Fixed! Try now.

Peter

[Peter Eastman]

Could you confirm whether the change I made on Feb. 5 fixed the problem for you?

Thanks!

Peter

[Kailong Mao]

Hi Peter:

Yes, now I can use "dt" for global computations! Thank you so much!