<div dir="ltr">Dear community, I am currently performing some all-electron geometry optimizations of W(OH)6 with CP2k to investigate the influence of relativistic effects. For the Tungsten atom, I use a basis set from Neese, which is specifically contracted for the use with DKH, and otherwise equals a TZVPP basis set ("SARC-DKH"). Without DKH switched on, everything runs fine. CP2k produces an optimized geometry which looks quite reasonable. Starting from this final geometry, I set up a second calculation, and just switched DKH3 on. Within the first steps of the optimization, everything looks fine: The hydroxyl groups all come a litter closer to the metal atom, exactly like expected from relativistic treatment. But then, one of the oxygen atoms comes closer and closer to the tungsten, much closer than it would be reasonable. The gradient is rising all the time, therefore these two atoms are attracting themselves even more the closer they penetrate. Now one could say that the BFGS geometry optimizer just performs strange/bad steps. But the strange thing is: The total energy is going down all of the time... In the end, when the atoms penetrate, the total energy is significantly lower than in the beginning, where the geometry was reasonable. Please find attached a zipped version of the calculation directory, including input file, log file, basis set and the trajectory of optimization (where you can see how the atoms penetrate). As already noted: Exactly the same input file, but without the &DKH section, worked fine. I appreciate any hints and suggestions concerning my problem. Best regards, Martin </div>