Prior distributions¶
Each free parameter of a retrieval calculations needs to have an associated prior distribution. The prior distribution describes any prior knowledge that is available about the parameter. The number and type of the free parameters depend on the chosen forward model and user-specified configuration options.
BeAR supports the following types of distributions:
delta
- a delta distribution with a fixed value
uniform
- a uniform distribution with a lower and upper bound
log_uniform
- a log-uniform distribution with a lower and upper bound
gaussian
- a Gaussian distribution with a mean and standard deviation
linked
- links this prior to that of another parameter
A special case is the linked
distribution. This distribution links the prior distribution
of one parameter to that of another. Thus, these two parameters will always have the same value during a retrieval
calculation.
Prior distributions units¶
BeAR also supports units for its prior distributions. The following units are currently taken into account:
Rs
orRsun
- the solar radius
Rj
orRjupiter
- Jupiter’s radius
Re
orRearth
- Earth’s radius
pc
- distance in parsec
ly
- distance in light years
Priors without units are assumed to be in cgs units.
Prior distributions file¶
The priors.config
file contains the information on the prior distributions of the free parameters.
The file has the following structure:
delta log_g 2.99
uniform r_planet 1.6 2.3 Rj
gaussian r_star 1.57 0.07 Rs
log_uniform H2O 1e-12 1e-1
log_uniform TiO 1e-12 1e-3
log_uniform VO 1e-12 1e-3
log_uniform K 1e-12 1e-3
uniform temperature 500 1600
The first column lists the distribution type of the prior, the second column the model parameter name, and the remaining columns the parameters of the distribution, while the optional, last column is the unit of the parameter.
The type, order and number of these parameters in prior distributions file is determined by chosen forward model.
A special case is the linked
distribution. This distribution links the prior distribution
of one parameter to that of another. The config parameter for this distribution is the line number of the
parameter distribution that it should be linked to.
An example of this is shown below:
uniform log_g 4.0 6.0
uniform scaling_factor 0.1 5.0
gaussian distance 5.8819 0.0029 pc
log_uniform mr_h2o 1e-12 0.01
log_uniform mr_ch4 1e-12 0.01
log_uniform mr_nh3 1e-12 0.01
log_uniform mr_k 1e-12 0.01
log_uniform mr_h2s 1e-12 0.01
linked mr_co2 5
uniform temperature1 5000 1000
uniform temperature2 0.3 0.95
uniform temperature3 0.3 0.95
uniform temperature4 0.4 0.95
uniform temperature5 0.5 0.95
Here, the prior for the CO2 mixing ratios is linked to the fifth parameter, which is the mixing ratio of CH4 in this example. Thus, CO2 will always have the same mixing ratio as methane for this retrieval setup. It is important to note that BeAR cannot check the consistency of the linked parameters. For example, if the linked parameter is a temperature, the resulting mixing ratios of CO2 would make no sense. It is the user’s responsibility to ensure that the linked parameters are consistent.
Order of prior distributions¶
The order of the prior distributions in the priors.config
file is very important. In general they have to appear in the
following order:
general forward model parameters, as discussed in the section on forward models
priors for the chosen chemistry models, as discussed in the section on chemistry models
temperature profile priors, as discussed in this section
cloud model priors as discussed in the section on cloud models
priors for optional modules that can be used for a specific forward model
observational offset priors as discussed in the section on observations
A general prior configuration file, therefore has the following structure:
uniform forward_model_param1 0.0 10.0
uniform forward_model_param2 0.0 10.0
uniform forward_model_param3 0.0 10.0
log_uniform chemistry_param1 1e-12 0.01
log_uniform chemistry_param2 1e-12 0.01
log_uniform chemistry_param3 1e-12 0.01
log_uniform chemistry_param4 1e-12 0.01
log_uniform chemistry_param5 1e-12 0.01
uniform temperature_param1 5000 1000
uniform temperature_param2 0.3 0.95
uniform temperature_param3 0.3 0.95
uniform temperature_param4 0.4 0.95
uniform temperature_param5 0.5 0.95
log_uniform cloud_param1 0.0 10.0
log_uniform cloud_param2 0.0 10.0
log_uniform cloud_param3 0.0 10.0
uniform module_param1 0.0 1.00
uniform module_param2 0.0 1.00
uniform module_param3 0.0 1.00
gaussian observation_offset1 100 0.4
gaussian observation_offset2 100 0.4