Identifying Halos in Cosmological Simulations with Continuous Wavelet Analysis: The 2D Case

TL;DR

This paper introduces a wavelet-based method for identifying halos in 2D cosmological simulation data, demonstrating its effectiveness, efficiency, and consistency with traditional methods, and highlighting its potential for large-scale data analysis.

Contribution

The study presents a novel 2D continuous wavelet transform approach for halo detection, offering smoother boundaries and better performance in sparse regions compared to traditional methods.

Findings

CWT-identified halos have smoother boundaries and are more compact.

The method operates with linear time complexity, suitable for large datasets.

Spatial distribution and power spectrum are consistent with traditional FOF halos.

Abstract

Continuous wavelet analysis is gaining popularity in science and engineering for its ability to analyze data across spatial and scale domains simultaneously. In this study, we introduce a wavelet-based method to identify halos and assess its feasibility in two-dimensional (2D) scenarios. We begin with the generation of four pseudo-2D datasets from the SIMBA dark matter simulation by compressing thin slices of three-dimensional (3D) data into 2D. We then calculate the continuous wavelet transform (CWT) directly from the particle distributions, identify local maxima that represent actual halos, and segment the CWT to delineate halo boundaries. A comparison with the traditional friends-of-friends (FOF) method shows that our CWT-identified halos, while contain slightly fewer particles, have smoother boundaries and are more compact in dense regions. In contrast, the CWT method can link particles over greater distances to form halos in sparse regions due to its spatial segmentation scheme. The spatial distribution and halo power spectrum of both CWT and FOF halos demonstrate substantial consistency, validating the 2D applicability of CWT for halo detection. Our identification scheme operates with a linear time complexity of $\mathcal{O}(N)$, suggesting its suitability for analyzing significantly larger datasets in the future.