An Algorithm to Compress Line-transition Data for Radiative-transfer Calculations

TL;DR

This paper introduces a temperature-dependent algorithm that compresses large molecular line-transition datasets for radiative-transfer calculations, significantly reducing data size while maintaining accuracy, thus enabling faster computations.

Contribution

The novel algorithm efficiently separates strong and weak line transitions, converting the majority into cross-section data and retaining detailed info for strong lines, improving computational feasibility.

Findings

Reduces line list size from 65 million to 7.7 million for HCN.

Maintains less than 1% difference in extinction spectra.

Enables faster radiative-transfer calculations with minimal accuracy loss.

Abstract

Molecular line-transition lists are an essential ingredient for radiative-transfer calculations. With recent databases now surpassing the billion-lines mark, handling them has become computationally prohibitive, due to both the required processing power and memory. Here I present a temperature-dependent algorithm to separate strong from weak line transitions, reformatting the large majority of the weaker lines into a cross-section data file, and retaining the detailed line-by-line information of the fewer strong lines. For any given molecule over the 0.3--30 {\micron} range, this algorithm reduces the number of lines to a few million, enabling faster radiative-transfer computations without a significant loss of information. The final compression rate depends on how densely populated is the spectrum. I validate this algorithm by comparing Exomol's HCN extinction-coefficient spectra between the complete (65 million line transitions) and compressed (7.7 million) line lists. Over the 0.6--33 {\micron} range, the average difference between extinction-coefficient values is less than 1\%. A Python/C implementation of this algorithm is open-source and available at https://github.com/pcubillos/repack . So far, this code handles the Exomol and HITRAN line-transition format.