Dear CP2K Developers, I am using REFTRAJ to compute the energy, forces, and stress for a trajectory of 100 structures sampled from a classical molecular dynamics simulation to build a training dataset. To validate the results, I compared the REFTRAJ outputs with those obtained from separate calculations on the first four structures. While the results are generally consistent, I noticed some small differences. For the first structure, the total energy differs at the 10th decimal place, the forces and stress are identical. These discrepancies increase slightly for subsequent structures. After examining the outputs from REFTRAJ and the separate calculations, I found that the energy differences from the second structure onward stem from variations in the Core Hamiltonian energy, Hartree energy, and exchange-correlation energy. I would like to understand what causes these differences and whether they are expected. Both methods use the same FORCE_EVAL settings, with SCF/SCF_GUESS set to ATOMIC and QS/EXTRAPOLATION set to USE_GUESS. The input scripts and output files are attached for your reference. Although these differences are too small to impact the quality of my training dataset, I am curious about the underlying reasons. I would appreciate any insights you can provide. Thank you in advance for your time and help! Best regards, Fengming Shi<div> </div><div><img alt="REFTRAJ_SP_results.png" width="535px" height="227px" src="cid:f6f35ee5-cb46-4066-bb37-875edfb4a125" /> </div> -- You received this message because you are subscribed to the Google Groups "cp2k" group. To unsubscribe from this group and stop receiving emails from it, send an email to <a href="mailto:cp2k+unsubscribe@googlegroups.com">cp2k+unsubscribe@googlegroups.com</a>. To view this discussion visit <a href="https://groups.google.com/d/msgid/cp2k/75867a5d-8eb1-431c-9f07-27639952a556n%40googlegroups.com?utm_medium=email&utm_source=footer">https://groups.google.com/d/msgid/cp2k/75867a5d-8eb1-431c-9f07-27639952a556n%40googlegroups.com</a>.