MCMC fitting problem

[danangsetyo5364]

Halo! Currently, I am trying to fit the light curve of detached binary with an MCMC sampler. My problem is, that I do not know why the solution does not fit with my data, despite the chain seems converging and staying at some parameters. I give my corner plot and the fitted result with burnin=1600 it, walker=150. Any idea how to fit it better? Thanks!

[kecnry]

It seems that some of the posterior distributions are "running up against a wall" of some sort. (Requiv ratio, F1, and F2, are all preferring lower values and do not have symmetric distributions. The mass ratio is relatively flat, which makes sense since it should not be well-constrained by a light curve and SB1 RVs.)

This can either happen from priors that you have assigned, parameter limits, or models that are unphysical or fail to compute. If you do have priors - check those first and see if they are reasonably set or if they should be removed/loosened.
If the priors are not to blame, then if you set expose_failed = True and then plot with style='failed', that might help to diagnose why the parameter space is not being explored in those regions. (See an example in Figure 9 here).

It might also be worth checking if the chain is actually flat and fully converged - how does the last plot look when you apply the same burnin = 1600?