If the computed equilibrium geometry appears to exhibit unlikely values, typically significantly too short bond lengths, you may have run into a basis set problem, in particular (but not only) if the Pauli relativistic method is applied.
Problem: Optimized bond lengths are clearly too short. The energy may also look suspicious.
Possible cause 1: Basis set trouble: onset of Pauli variational collapse, if you have applied the Pauli relativistic option. Caused by small (or absent) frozen cores and/or relatively large basis sets, applied to heavy elements.
Possible cause 2: Basis set trouble also, but quite different from the previous potential cause: you have used relatively large frozen cores. When the atoms approach each other during the optimization and the frozen cores start to overlap, the energy computation and the computed energy gradients become more and more incorrect. This is a result of the inappropriateness of the frozen core approximation, which indeed assumes that frozen cores of neighboring atoms do not significantly overlap. Without going into a detailed explanation here, the net effect is that certain repulsive terms in the energy computation are missing and hence a spurious tendency to a 'core collapse' arises, yielding too short bond lengths.
Cure: Best is to abandon the Pauli method and use the ZORA approach instead for any relativistic calculation. If for whatever reason you insist on using the Pauli formalism, apply bigger frozen cores and, if that doesn't help, reduce the basis set (not by deleting polarization functions, but by reducing the flexibility of the occupied-atomic-orbitals space, in particular s- and p-functions). Note, however, that large frozen cores can be a cause for trouble by themselves, irrespective of any relativistic feature. If you have reason to believe that your frozen cores might be too large, given the resulting bond lengths in your calculation, you have to pick smaller cores (and hence be very wary of using the Pauli formalism for any relativity).
