Efficient, uninformative sampling of limb darkening coefficients for two-parameter laws

TL;DR

This paper introduces an efficient sampling method for limb darkening coefficients in two-parameter laws, ensuring physical plausibility and reducing parameter correlations, thereby improving the robustness of stellar intensity profile modeling.

Contribution

It presents a novel, optimal sampling technique using triangular sampling and Dirichlet distributions for two-parameter limb darkening laws, with practical parametrizations for common models.

Findings

Triangular sampling efficiently generates physical limb darkening coefficients.

The new parametrizations reduce mutual correlations among parameters.

The approach yields more robust and realistic uncertainty estimates.

Abstract

Stellar limb darkening affects a wide range of astronomical measurements and is frequently modelled with a parametric model using polynomials in the cosine of the angle between the line of sight and the emergent intensity. Two-parameter laws are particularly popular for cases where one wishes to fit freely for the limb darkening coefficients (i.e. an uninformative prior) due to the compact prior volume and the fact that more complex models rarely obtain unique solutions with present data. In such cases, we show that the two limb darkening coefficients are constrained by three physical boundary conditions, describing a triangular region in the two-dimensional parameter space. We show that uniformly distributed samples may be drawn from this region with optimal efficiency by a technique developed by computer graphical programming: triangular sampling. Alternatively, one can make draws using a uniform, bivariate Dirichlet distribution. We provide simple expressions for these parametrizations for both techniques applied to the case of quadratic, square-root and logarithmic limb darkening laws. For example, in the case of the popular quadratic law, we advocate fitting for q_1 = (u_1+u_2)^2 and q_2 = 0.5u_1(u_1+u_2)^{-1} with uniform priors in the interval [0,1] to implement triangular sampling easily. Employing these parametrizations allows one to derive model parameters which fully account for our ignorance about the intensity profile, yet never explore unphysical solutions, yielding robust and realistic uncertainty estimates. Furthermore, in the case of triangular sampling with the quadratic law, our parametrization leads to significantly reduced mutual correlations and provides an alternative geometric explanation as to why naively fitting the quadratic limb darkening coefficients precipitates strong correlations in the first place.