Generating on-the-fly large samples of theoretical spectra through N-dimensional grid

TL;DR

This paper presents a method to efficiently generate large, on-the-fly theoretical spectra grids in multiple dimensions for astronomy, enabling improved parameter estimation with reduced storage and update costs.

Contribution

It introduces an N-dimensional grid construction and interpolation scheme for generating spectra, which is adaptable, storage-efficient, and improves parameter estimation accuracy.

Findings

Mean square errors decrease with grid resolution.

The scheme is applicable to other multi-dimensional modeling studies.

Efficient updating of the spectra grid is possible.

Abstract

Many analyses and parameter estimations undertaken in astronomy require a large set (> 10^5) of non-analytical, theoretical spectra, each of these defined by multiple parameters. We describe the construction of an N-dimensional grid which is suitable for generating such spectra. The theoretical spectra are designed to correspond to a targeted parameter grid but otherwise to random positions in the parameter space, and they are interpolated on-the-fly through a pre-calculated grid of spectra. The initial grid is designed to be relatively low in parameter resolution and small in occupied hard disk space and therefore can be updated efficiently when a new model is desired. In a pilot study of stellar population synthesis of galaxies, the mean square errors on the estimated parameters are found to decrease with the targeted grid resolution. This scheme of generating a large model grid is general for other areas of studies, particularly if they are based on multi-dimensional parameter space and are focused on contrasting model differences.