<div dir="ltr">Dear CP2K users and developers, I would like to perform a MD simulation for a transition metal complex in water; the complex being in an open-shell state. However, I'm facing an issue which I cannot resolve: the time taken by a MD step dramatically increases after some steps because of the increase of the number of SCF cycles (see Figure_1.eps). I first thought that this was due to the order chosen for the wavefunction extrapolation (ASPC 3), but the situation does not change when varying the order (1-4). I usually tend to use rather conservative values for EPS_DEFAULT and CUTOFF (1e-14 and 1000 Ry), and I was wondering if this could lead to a numerical instability in the present case. However increasing EPS_DEFAULT to 1e-12 does not improve the situation; likewise for using a smaller CUTOFF value of 500 Ry. Please note that, when the simulation is restarted, the same phenomenon in observed. Thus, in Figure_2.eps, the simulation was restarted at ~1070 fs. I made several attempts to find which mistake(s) I could have made (CG vs DIIS minimizer; EPS_SCF value of 1e-6 or 5e-7; CSVR versus NHC thermostating; TIMESTEP of 0.25 or 0.50 fs; PRECONDITIONER and ENERGY_GAP ...): but nothing came out of this. There is perhaps a trivial error which I may be missing. Please let me know: I have attached the input file to the message. Thanks in advance for your help. Best regards, Max P.S. For Figure_1.eps, the simulation was perfomed using the attached input file. For Figure_2.eps, there are the following changes: "EPS_DEFAULT 1.0E-12", "EXTRAPOLATION_ORDER 1", "PRECONDITIONER FULL_SINGLE_INVERSE". </div>