Replies: 1 comment 1 reply
-
We've actually been looking into this already this week and @kmhambleton opened #901 to change the behavior for RVs. Are the changes there in agreement with what you'd expect? |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hello,
I'm using PHOEBE to fit a light curve and RV dataset of a detached EB. My RV uncertainties are likely underestimated, so I'd like to sample over sigmas_lnf when I run MCMC.
I am confused about the implementation of sigmas_lnf in PHOEBE. The documentation suggests that the sigmas are increased by
interpolated_model**2 * np.exp(2 * sigmas_lnf)
Why does this term include the interpolated model? For example, if I have two identical systems and one has vgamma=10km/s and the other has vgamma=100 km/s, why would the RV uncertainties be rescaled to larger values in the later case? I would also think that the RV observations should know nothing about the phase of the binary at the observation time.
Reading through the emcee line fitting tutorial that is linked in the emcee sampler example, it seems like their likelihood function adds
model**2 * np.exp(2 * sigmas_lnf)
to the "measured" uncertainties because they've intentionally set up an example where the uncertainties are underestimated more at larger y values.How would you recommend using the sigmas_lnf parameter for an RV dataset?
--Dom
Beta Was this translation helpful? Give feedback.
All reactions