<div dir="ltr">I'd have to agree, TYPE GEO_OPT can be a terribly inefficient use of resources. If the geometry in the cell is so poor you need to optimize it before relaxing the cell dimensions, just do a RUNTYPE GEO_OPT to tidy things up and follow up with RUNTYPE CELL_OPT and TYPE DIRECT_CELL_OPT. Or run a short annealing simulation. Sometimes that's even better because you'll generate a distribution of starting points to find the minimum. To Matthias' point, I'm currently using a version of CP2K 5.0 I downloaded a couple months ago. Switching from TYPE GEO_OPT to TYPE DIRECT_CELL_OPT, I was unable to reproduce your issue. Cell angles stayed at 90 degrees. I made only one other change to the second input file you provided and that was to reduce REL_CUTOFF to 40 Ry to make decent progress on a single node. I killed the job after 15 minutes. Here's the printout before beginning step 50... -------------------------- OPTIMIZATION STEP: 50 -------------------------- CELL| Volume [angstrom^3]: 91.561 CELL| Vector a [angstrom]: 4.286 0.000 0.000 |a| = 4.286 CELL| Vector b [angstrom]: 0.000 4.787 0.000 |b| = 4.787 CELL| Vector c [angstrom]: 0.000 0.000 4.462 |c| = 4.462 CELL| Angle (b,c), alpha [degree]: 90.000 CELL| Angle (a,c), beta [degree]: 90.000 CELL| Angle (a,b), gamma [degree]: 90.000 CELL| Numerically orthorhombic: YES -Travis On Tuesday, October 31, 2017 at 5:51:56 AM UTC-4, Tobias Binninger wrote:<blockquote class="gmail_quote" style="margin: 0;margin-left: 0.8ex;border-left: 1px #ccc solid;padding-left: 1ex;"><div dir="ltr">Hi Matthias, thanks, and thanks a lot for the hint regarding cell optimization. Best regards, Tobias </div></blockquote></div>