Thermostat regions in MD

[Philippe C. A.]

Hi all,

I have a simple MM-only system of a couple hundred atoms of types A
and B. What I would like to do: Attach a Nose-Hoover or Gaussian
Isokinetic thermostat to all atoms of type B, but not A, i.e.
particles B move according to the Nose-Hoover (or GIK) equations of
motion, whereas particles A have no thermostat and simply move
according to Newtonian dynamics.

>From the input manual I had some hope that this would be possible
using the "thermal_region" and "define_region" subsections and from my
interpretation of it I created an input structure like this and
thought it would do what I want:

&MOTION
 &MD
  ENSEMBLE NVT
  TEMPERATURE [K] 500.0
  &THERMOSTAT
   TYPE NOSE
   REGION DEFINED
   &DEFINE_REGION
    LIST 1 2 3 [...] 250   # My system has 250 particles A and 250
particles B
   &END DEFINE_REGION
   &NOSE
    LENGTH 1
    TIMECON [fs] 1.0e-2
   &END NOSE
  &END THERMOSTAT
  TIMESTEP [fs] 1.0 fs
  STEPS 20
 &END MD
&END MOTION

The result however was disaster - all NaN positions/forces and CP2K
complains that "no thermostat is attached to particles 251-500" (and
if I do try to attach one to them, it complains that the thermostat
input section cannot be repeated....).
With the GIK thermostat - which in CP2K is not treated as a thermostat
but as an ensemble - I tried the "motion%md%thermal_region" subsection
in the hope that it would only treat the listed atoms within the
ISOKIN ensemble and the non-listed atoms with simply Newtonian EOM,
but that does not happen.

Could somebody clarify the intended use of the "region" keywords and
how they work, and if it is possible to achieve what I described above
with the current CP2K functionality with Nose and Gaussian
thermostats?

Many thanks,
Philippe

PS: I am using the "release" branch of CP2K.