Improved two-point correlation function estimates using glass-like distributions as a reference sample

• Published in: Monthly notices of the Royal Astronomical Society Vol. 506; no. 4; pp. 4667 - 4675
• Main Authors: Dávila-Kurbán, Federico, Sánchez, Ariel G, Lares, Marcelo, Ruiz, Andrés N
• Format: Journal Article
• Language: English
• Published: Oxford University Press 01-10-2021
• Subjects: methods: analytical; methods: statistical; galaxies: statistics; large-scale structure of Universe
• ISSN: 0035-8711; 1365-2966
• DOI: 10.1093/mnras/stab1622

ABSTRACT
All estimators of the two-point correlation function are based on a random catalogue, a set of points with no intrinsic clustering following the selection function of a survey. High-accuracy estimates require the use of large random catalogues, which imply a high computational cost. We propose to replace the standard random catalogues by glass-like point distributions or glass catalogues whose power spectrum P(k) ∝ k4 exhibits significantly less power on scales larger than the mean interparticle separation than a Poisson distribution with the same number of points. We show that these distributions can be obtained by iteratively applying the technique of Zeldovich reconstruction commonly used in studies of baryon acoustic oscillations (BAO). We provide a modified version of the widely used Landy–Szalay estimator of the correlation function adapted to the use of glass catalogues and compare its performance with the results obtained using random samples. Our results show that glass-like samples do not add any bias with respect to the results obtained using Poisson distributions. On scales larger than the mean interparticle separation of the glass catalogues, the modified estimator leads to a significant reduction of the variance of the Legendre multipoles ξℓ(s) with respect to the standard Landy–Szalay results with the same number of points. The size of the glass catalogue required to achieve a given accuracy in the correlation function is significantly smaller than when using random samples. Their use could help us to drastically reduce the computational cost of configuration-space clustering analysis of future surveys while maintaining high-accuracy requirements.