Please “reply to all” to keep the list in the loop.
I forgot about FIRE not working for variable boxes. Yes, minimization in systems that use reaxff has proven challenging. If you are trying to reproduce the results of a certain paper perhaps it is best to set the exact lattice they report and use the iso option instead of aniso. You could also try relaxing first just the atomic positions and then both box and positions. A problem could be that your particular reaxff uses control parameters different from the default hard-coded values within Lammps. And don’t use force tolerances of 1e-10 for a FF that has discontinuities like reaxff. Start with something more reasonable (1e-2) where convergence should take less time thus allowing to troubleshoot more efficiently and lower it as you go. You could also run damped dynamics and look at the time evolution of the total energy to better understand what’s happening during relaxation.
Carlos