<div class="gmail_quote"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">no, a thermostat can hide a lot of crap. if you cannot conserve energy in an NVE ensemble well enough you are in trouble. i found that already 0.5fs timestep was an agressive choice for flexible water and 0.25fs a conservative choice. 1fs would be very aggresive for deuterated water. (0.7fs would be better). adding shake can allow up to 2fs, 2.5 if you are very aggressive with classical potentials. </blockquote></div> Thanks! This is something useful! Now I know why classical simulations in the literatures were using 2 fs. Maybe the project is different, I would like to say 0.5 fs for H2O should be fine enough (see Fig.5 in CPC 167 (2005)103) and 1 fs can obtain reasonable results. In addition, I have checked the stand output and found *SCF run converged in n steps * except for the first step, which need 30 + 9 steps. Here "n" is typically around 9. At the very beginning I also compared some QS tests from between different machines. They are digitally the same. I guess that the time consuming is time dependent may come from another reason. I'll keep it in mind and try to look at more examples and keep you informed. Sincerely, Yunfeng Sincerely, Yunfeng