<div dir="ltr">Dear Samuel,<br><br>thank you for your suggestions. I will try first without curvy_steps section and see what the result is. After that I start with curvy_steps (after deleting TRS4).<br><br>Unfortunately our super computer system does not still support OpenMPI fully, but there are some newly updated versions of MPI for faster parallel computations.<br>So I have to write the code with "srun" and set the number of nodes and tasks-per-node as high as possible. Let's see how fast and effective it works. After finishing a part of the simulation I will give an overview of performance results here.<br><br>Kind regards,<br><br>Phil<br><br>Am Freitag, 16. September 2016 11:23:45 UTC+2 schrieb Samuel Andermatt:<blockquote class="gmail_quote" style="margin: 0;margin-left: 0.8ex;border-left: 1px #ccc solid;padding-left: 1ex;"><div dir="ltr">OPTION 1 (more stable):<br>delete PURIFICATION_METHOD TRS4<br>OPTION 2 (probably better performance):<br>delete curvy_steps section<br>add <a href="https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/LS_SCF.html#list_MIXING_FRACTION" target="_blank" rel="nofollow" onmousedown="this.href='https://www.google.com/url?q\x3dhttps%3A%2F%2Fmanual.cp2k.org%2Ftrunk%2FCP2K_INPUT%2FFORCE_EVAL%2FDFT%2FLS_SCF.html%23list_MIXING_FRACTION\x26sa\x3dD\x26sntz\x3d1\x26usg\x3dAFQjCNGGTDroUDy1iqFD9cAx_JVHQ9ZKYA';return true;" onclick="this.href='https://www.google.com/url?q\x3dhttps%3A%2F%2Fmanual.cp2k.org%2Ftrunk%2FCP2K_INPUT%2FFORCE_EVAL%2FDFT%2FLS_SCF.html%23list_MIXING_FRACTION\x26sa\x3dD\x26sntz\x3d1\x26usg\x3dAFQjCNGGTDroUDy1iqFD9cAx_JVHQ9ZKYA';return true;">MIXING_FRACTION</a> 0.2<br><br>I would try both out and see which one can converge and which is faster.<br></div></blockquote></div>