Orbital Transformation in MD

[August Melcher]

Hello all,

I've been running molecular dynamics simulations using the orbital 
transformation method and the diagonalization method and I've noticed that 
while orbital transformation takes much less CPU time, there are timesteps 
where the CPU time deviates from the average by a huge margin (~3500s for 
one particular step vs ~20s average/step in a simulation I recently ran). 
Does anyone have any useful insight into why this is happening and possibly 
how it could be avoided?

Thanks,
August
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140407/3b721a90/attachment.htm>