1) Switch on force calculation in dmft2 step. 
  Change the flag "TOT" in case.in2 file to "FOR"
2) Increase number of DFT steps in "params.dat" file, so that DFT is basically converged at each self-energy update.
  Change "max_lda_iterations=100" in "params.dat"
3) Finally, switch on structure optimization by changing the flag "MSR1" in case.inm to "MSR1a"

Note that it is more efficient to first converge the DMFT calculation
[using "MSR1" mixer setting and "max_lda_iterations=10"], and once
self-energy stops changing, one can change the two parameters to
["MSR1a" and "max_lda_iterations=100"].  

Alternatively, one can resubmit a converged DMFT run for
optimization. In this case, it is a good idea to set "runIMP=False" at
the very beginning, so that structure is optimized on well converged
self-energy. During the job is going through the first iteration, one
should switch "runIMP=True", so that self-energy is updated in the
second iteration.

