Using a negative 'dt' in the mdp file does not not seem to cause any
simulation problems (tested on a waterbox- doesn't crash). I am not sure
if the system is evolving backwards in time though. I am not sure how to
check that.  

But I think, all I require is that the equations of motion be integrated
backwards. The ensemble of trajectories which will end up in one  state A 
(an attractor) when integrated backwards and in state B (another
attractor) when integrated forwards will be the ensemble I need for
transition path sampling. 

Sumanth N Jamadagni
Graduate Student 
Isermann Dept of Chemical and Biological Engg
Rensselaer Polytechnic Institute

jamads@rpi.edu
(Cell)518-598-2786