Platon

This corresponds most closely to the version described in Zhang et al 2024. It is highly recommended to upgrade to this version. Major changes include:

- GPU support through cupy. If cupy is not detected, PLATON will use the CPU
- Free retrievals with VMR and CLR priors
- Higher-resolution (R=20k) and more up-to-date opacities, which also cover higher temperatures (up to 4000 K) and bluer wavelengths (down to 200 nm)
- Pareto importance sampling leave-one-out cross-validation
- Surface emission with albedos from real lab data (beta), courtesy of Kim Paragas
- Support for pymultinest in addition to dynesty, which is often much faster
- Slightly better definitions of metallicity and C/O ratio
- Support for a variety of Mie scattering hazes, with wavelength-dependent complex refractive indices from the literature
- Much better plotting tools, courtesy of Yayaati Chachan

This is the PLATON release that corresponds to the final version of the second PLATON paper (https://arxiv.org/abs/2004.09513).

Improvements:

1. Example files are a lot better
2. final evidence is made more explicit, stored in final_logz of the RetrievalResult object
3. eclipse_depth_calculator now uses Simpson's rule instead of Trapezoid rule, which improves accuracy by ~1%
4. Collisional data initialized in constructor, which speeds up R=10,000 calculations
5. Added warnings for when the user species bins that are too small, or tries to use high resolution opacities without specifying that the star is a blackbody
6. The radiative transfer calculation for emission is now reformulated to use E3 instead of E2, which guarantees exactly the right result for an isothermal atmosphere, and better error properties for near-isothermal atmospheres.

Bug fixes:

1. RetrievalResult plot_specturm now handles correlated k results properly
2. For correlated k, unbinned_wavelengths, unbinned_depths, etc now work properly

The attached zip file is the data directory associated with these release. It should not be necessary because PLATON automatically downloads the data from a server. We upload it here in case the server ever goes down.

paper_II
This is the version that corresponds most closely to the draft version of the second PLATON paper (https://arxiv.org/abs/2004.09513), but we recommend you use v5.2.1 instead, which corresponds to the final version of the paper. We created the release in order to obtain a DOI with Zenodo.

The attached zip file is the data directory associated with these release. It should not be necessary because PLATON automatically downloads the data from a server. We upload it here in case the server ever goes down.

PLATON v5 is a major release, corresponding to the version described in our second PLATON paper (submitted). If you cloned the repository before 02/02/2020, PLEASE RE-CLONE THE REPOSITORY AND DELETE THE OLD ONE, due to breaking change 2 below. The most important changes are:

Breaking changes:

1. Removed data from repository and its history. Data is now stored elsewhere, and will be automatically downloaded on the first initialization of the transit or eclipse calculators. If you cloned the repository before 02/02/2020, PLEASE RE-CLONE THE REPOSITORY AND DELETE THE OLD ONE to avoid data corruption.
2. run_multinest and run_emcee now return RetrievalResult object, which contains observational data, fit info (fixed parameters, fit parameters, fit priors), best fit spectrum, ancillary data for the best fit spectrum (T/P profile, abundances, stellar spectrum, unbinned best fit spectrum, unbinned stellar spectrum, unbinned correction factors, etc), and 100 randomly sampled unbinned spectra (to plot 1 sigma range). The RetrievalResult has two methods: plot_spectrum() and plot_corner()
3. Retriever deprecated. Please use CombinedRetriever instead
4. Removed support for Python 2.

Significant new features:
1. Correlated k support added
2. Buggy opacities for C2H2, C2H6, HF, N2, NO2, O2, O3, OCS have been fixed. NH3 opacity now uses the CoYuTe line list, while CH4 opacity uses Rey et al 2017
3. H- opacity now supported, though disabled by default because it's rarely important in the temperature range we support
4. A quench pressure is now implemented

Other improvements:
1. BT-NextGen stellar grid replaced with BT-Settl (better for cooler stars)
2. Enabling wavelength-dependent cross sections for MgSiO3, TiO2, and SiO2
3. Retrieval: complex refractive index split into two components, n and log_k, so that the refractive index can be retrieved.
4. Added wfc3_transit_offset and wfc3_eclipse_offset as retrieval parameters
5. Can now use blackbody spectrum for star
6. More accurate interpolation method of absorption cross sections

Improvements:
1. Switching from nestle to dynesty (BREAKING CHANGE because retriever returns different object)
2. Nested sampling now saves two diagnostic plots: runplot and traceplot (see dynesty documentation for an explanation of these plots)
3. Eclipse depth calculator now much faster due to use of special function expn
4. Radiative solution T/P profile now accepts log(kappa_IR) and log(gamma) instead of the un-logged variables. This is because it is better for retrievals to use log-uniform rather than uniform priors.
5. Number of heights in profile dropped from 500 to 250 for speed
6. Added logo

Bug fixes:
1. Radiative solution T/P parameterization's alpha is changed to 1-alpha to be consistent with Line et al 2013
2. Corrected bug in blackbody equation for stellar spectra. Does not apply to cases where we use the PHOENIX spectrum (which should be most cases)

Bug fixes:

1. Multinesting previously used ln_prob while nestle expected ln_like, causing posteriors to be narrower and closer to the prior for parameters with Gaussian priors.
2. For multinesting, the previously reported best fit was the fit with the highest likelihood, not the one with the highest posterior probability. The latter is correct.
3. set_from_arrays in T/P profile was previously broken

Improvements:
1. Fixed problem in opacities for HITRAN molecules (C2H2, C2H6, CO2, HF, N2, NO2, O2, O3, OCS) plus carbon monoxide where the Voigt cutoff was 25 cm^-1 even for high pressures. It is now 100 cm^-1 for 10 and 100 bar, and 1000 cm^-1 for 1000 bar. Also, HITRAN opacities now include all isotopologues.
2. Na and K opacities are now calculated from NIST line data.
3. The radius used in eclipse depth calculations is now the radius where the radial optical depth is 1. Previously, it was the radius where P = 1 bar.
4. Unbinned transit and eclipse depths are now plotted with alpha=0.2, to prevent blocking everything else from view

This is a major update on v2, the version our paper is based on. We made substantial improvements to PLATON:

1. Eclipse depth calculator is now out of beta. It now supports GPU acceleration on CUDA-capable devices: just install CUDA, cudamat, and gnumpy, and PLATON should automatically detect gnumpy

2. All molecular absorption data has been regenerated using the latest line lists from ExoMol (most molecules), HITRAN (some molecules), CDSD-4000 (CO2), and Schwenke (TiO)

3. v3 is more than two times faster than v2. The speed improvement comes from two places: a) we transposed data arrays to improve cache locality, which speeds the code up by 1/3; b) reduced the number of layers the atmosphere is divided into from 500 to 250, which changes transit depths by less than a percent but substantially speeds up the code