Hi,
I've rerun the simulations with 'random clcg' invoked in the script and this gets rid of the error in the correlation function (at least from what I've simulated up to 1.5*10^6 ps of simulation time). So it was the short period of the old random number generator that caused the original problem.
Thanks for your help Rick and Lennart. I'm glad to hear that the new random number generator will be default and not need to be invoked explicitly in future releases of CHARMM.
Best wishes,
Ed