What is kima

Kima is the name of a (really good!) drink produced in Azores.
And now, kima is also a package for the analysis of radial velocity (RV) data.

---

kima fits Keplerian curves to a set of RV measurements, using the Diffusive Nested Sampling algorithm (Brewer et al. 2011) to sample the posterior distribution for the model parameters. Additionally, kima can calculate the fully marginalized likelihood (or evidence) of a model with a given number of Keplerians and also infer the number (Np) of Keplerian signals detected in a given dataset. For this latter task, Np is a free parameter in the fit and is inferred using the methods proposed by Brewer (2014) and Brewer & Donovan (2015).

Cool, I want to try it!

Credit

If you make use of this code, please cite the following paper:

Faria et al., (2018).
kima: Exoplanet detection in radial velocities. Journal of Open Source Software, 3(26), 487
https://doi.org/10.21105/joss.00487

@article{kima,
   author = {Faria, J. P. and Santos, N. C. and Figueira, P. and Brewer, B. J.},
   title = {{kima}: Exoplanet detection in radial velocities},
   journal = {Journal of Open Source Software},
   year = 2018,
   volume = 3,
   number = 26,
   pages = 487,
   doi = {10.21105/joss.00487},
   url = {https://doi.org/10.21105/joss.00487}
}

Doesn't [insert code] do this already?

One can very easily find a large number of other packages (e.g. radvel, exonailer, EXOFAST, pyaneti, ktransit, PyOrbit) aimed at the analysis of RV data, and which implement a subset of kima's features or vice versa. So, take the following into consideration:

advantages

• kima allows for the number of Keplerian signals to be inferred together with the other parameters (to our knowledge no other open-source package except Exoplanet, on which kima is based, allows for this)
• kima uses Diffusive Nested Sampling to sample the parameter posteriors, which means
  • it's less prone to "getting stuck" in local likelihood maxima
  • the sampling is still efficient in multimodal posteriors
  • the evidence is obtained from a single run
  • no need for initial guesses for the parameters (!)
• kima can easily use a correlated noise model (based on Gaussian processes) for stellar-induced signals
• kima is written in efficient and multi-threaded C++

disadvantages

• kima does not consider photometric (transit) lightcurves
• there are probably bugs 🐛 lurking around, be careful and contribute if you catch one!