Hello Mahbub, please write to the mailing list for quick questions regarding the software and simulation protocol. Then if you have other, non-quick question, we can discuss them again.
The input looks mostly OK, except that the correlation function calculation is meaningless when you’re adding a biasing potential.
Strictly speaking, umbrella sampling itself doesn’t give you a PMF, because it is only a method to enhance sampling through an entire region of phase space.
You will need a method to remove the effects of the harmonic bias, which you can calculate because you added the bias, to extrapolate a probability distribution under unbiased conditions. This is done most often by the Weighted Histogram Analysis Method (WHAM), for which the most common implementation is:
More recently, people have used various types of bootstrapping techniques, which you can also play with.
After getting a probability distribution for z, the free energy is simply its logarithm (WHAM does this for you already).