Dear Kunsa Haho , I had the interesting answer from ChatGPT. I hope that it will be usefull for your research:

"One way to overcome this problem is to use a larger convergence threshold for the ionic relaxation. You can do this by adjusting the input parameters for the vc-relax calculation in Quantum ESPRESSO. Another way is to use a larger value for the electronic damping factor, which can help stabilize the ionic positions during the relaxation. Additionally, you can try using a different type of ionic relaxation algorithm, such as conjugate gradient or Broyden-Fletcher-Goldfarb-Shanno (BFGS) method. It is also possible that the atoms are not being missed but are instead being moved to a different location in the structure. To confirm this, you can check the final coordinates of the atoms in the relaxed structure."

The atoms aren't gone, this is just a difference attributable to your visualizer. The purple/red atoms in the left panel are periodic images of one another, they're just not drawn in on the right panel because the vc-relax took the distance between purple/silver atoms to a point where they're > than the visualizer's in-built bond cutoff.