Hi, I tried some test to assess the suggestions from Jorg and Matt. It follows the results. First the Matt's suggestions (all_matt_50ps.png): - only decreasing the time step to 2.0 fs don't change the effect (red lines in NVT and NVE runs); - putting the SHAKE_TOLERANCE key allows to get better results in the NVT runs but in the NVE ones it is appreciable a drift (blue and green lines). However, the outcome is that the results with time step 2.0 and 2.5 are quite indistinguishable. Regard to Jorg's suggestions (all_jorg_50ps.png): - only changing the ewald parameter we get a drift even larger (red lines); - by combining ewald changes with an enlarged shake tolerance I get the correct behaviour (blue lines). To confirm it, I performed also a run with flexible molecules (all_flexible_50ps_cut.png) and I got almost flat profiles. I tried also to change the real space cutoff value (that is not the same of the LJ interactions) but I got identical results as the one that I showed. Best Paolo Il giorno martedÃ¬ 25 settembre 2012 11:45:03 UTC+2, paolo nicolini ha scritto:<blockquote class="gmail_quote" style="margin: 0;margin-left: 0.8ex;border-left: 1px #ccc solid;padding-left: 1ex;">Hi thank you for the quick answers! I will try to follow your suggestions... My version is: CP2K| version string: CP2K version 2.4 (Development Version) CP2K| source code revision number: 12363 CP2K| is freely available from <a href="http://www.cp2k.org/" target="_blank">http://www.cp2k.org/</a> CP2K| Program compiled at Fri Sep 7 12:38:53 CEST 2012 CP2K| Program compiled on titania1 CP2K| Program compiled for Linux-i686-intel Best Paolo </blockquote>