Hi Jun
then you don't need at all the different thermostatting region at
different temperatures.
You can specify velocities in the section &VELOCITIES in &SUBSYS.
These velocities act like restart quantities.
You can (with a small external program) determine the velocities of the
hotter atoms through a Boltzman distribution at temperature T1, and the
velocities of the other atoms at temperature T2 (T1>T2).
You can then input these quantities in the &VELOCITIES section and run
your MD.
Of course you may consider about running an NVE ensemble instead of the
NVT (which you were mentioning in your previous message). NVE makes
more sense to me in order to inspect heat transfer (but of course more
expert people can comment on this point).
Cheers
Teo