We describe a general engineering method to compute a tighter bound on fiber-optic channel capacity and the optimum source distribution. This method first determines the conditional probability density functions of an optical fiber channel by evaluating histograms, or by instantons or edgeworth expansion approach. Starting from any source distribution, we generate source and output sequences to estimate the information rate and calculate A-posteriori State-Transition Weight. The optimum source distribution is obtained by constrained stochastic Arimoto-Blahut algorithm. We apply the proposed methods on a 64 iterative polarization quantization source. The optimized capacity per single-polarization per channel is higher than achievable information rate. As the transmission distance increases the optimized distribution is changed from Gaussian-like distribution to non-Gaussian-like distribution. This is due to the fact that the nonlinear interaction of ASE noise and Kerr nonlinearities cannot be compensated for by back-propagation method.