Rule of thumb:

If the energy is the same for the first hundred steps or so, and (if the numbers are not close to 0) diverge only by on the order of 1 in the 3rd significant figure after 1000 steps, the results are exactly the same, modulo machine precision; slight differences in the ordering of operations by compilers and different roundoff procedures used by different hardware lead to slight differences accumulated over many cycles.

This ordinarily leads to different numbers of steps before convergence in energy minimization.

If the numbers are small, this effect can be exaggerated, but average properties (in dynamics) are likely to agree better.

Example: final state of energy minimization.
`
`

NSTEP ENERGY RMS GMAX NAME NUMBER 13013 542.823 0.000 0.001 C5 396 BOND = 874.5219 ANGLE = 108.3989 DIHED = 177.3000 VDWAALS = 177.8272 ELECT = 358.5989 HBOND = -4.0348 1-4 NB = 80.8837 1-4 ELEC= -1230.6732 CONST = 0.0000 --- NSTEP ENERGY RMS GMAX NAME NUMBER 12009 542.823 0.000 0.001 C13 94 BOND = 874.5233 ANGLE = 108.4039 DIHED = 177.2966 VDWAALS = 177.8211 ELECT = 358.6016 HBOND = -4.0350 1-4 NB = 80.8840 1-4 ELEC= -1230.6729 CONST = 0.0000