How to Tackle the Computational Challenges of Line-by-line Modelling

Abstract

Despite of significant advances of computational power, line-by-line modeling of molecular absorption and radiative transfer remains a challenge. In case of Earth atmospheric remote sensing, the number of spectra to be processed operationally has grown substantially. In contrast the number of planetary spectra is quite limited, however, the number of spectroscopic transitions to be considered, esp. for hot Jupiters (or Venus in our solar system), has increased to millions or billions. In this contribution we present our approach to tackle this computational challenge in our lbl atmospheric radiative transfer code GARLIC (Generic Atmospheric Radiation Line-by-line Infrared Code, Fortran 2003). A highly optimized Voigt function algorithm along with a 'multigrid' approach is employed for computation of lbl molecular absorption cross sections. Multithreading by OpenMp allows to further speed-up the cross section, absorption coefficient, and atmospheric radiative transfer modules. Furthermore, we briefly discuss the option of hardware accelerators for the most demanding cross sections. Finally, Jacobians are required to solve the nonlinear inverse problems to be solved for the analysis of an increasing number of exoplanet observations and we present algorithmic differentiation as a technique to implement accurate and efficient derivatives.