You are correct. Both comb and reaxff use some variation of the Rappe/Goddard or Stuart/Rick methods for charge equilibration. In this method, the charge on each atom is treated as a dynamic variable which is determined using Lagrangian mechanics in exactly in the same way traditional MD determines the atomic positions. Basically you have one equation of motion for both particles and charges. An appropriate choice of the charge fictitious mass makes this possible. This makes for a relatively inexpensive calculation.
To minimize energy of the system, you would need to rewrite the charge equilibration in such a way that at every energy calculation the charge is determined self-consistantly each time you perturb the atomic nuclei positions. Essentially you’d be using a Born-Oppenhiemer type approach… This approach would be very expensive for MD, but would work for energy minimization.
At present, I don’t think anybody has a code to do this. Best bet would be to do a slow simulated anneal from relatively high temperature to low temperature, and hope for the best.
PS: Xiaoyin Ji say hello to Don Brenner and Doug Irving for me!