So let’s go by parts MCMC and PyMC3 is a smart sampler that uses your parameter priors to build Markov chains that contain your posterior distributions for each parameter.
This being said, you can always update your priors for example for Ea_r you can use in a second step something like Uniform(3,7) instead of your initial prior… Why I am saying this, for you to force the sampler to draw samples in a reduced region.
In order to check if the mcmc samples have converged you can always plot the autocorrelation (good case is near 0) something like this:
pm.plot_autocorr(trace)
Hope my advice helps you…