Machine-learning classification of X-ray sources

The rapid increase in serendipitous X-ray source detections requires novel approaches to efficiently classify them. Only a small fraction of detected sources have reliable identifications — for example, roughly one quarter of the ~700 X-ray sources in M33. I am developing an automated classification pipeline (XClass) using supervised machine-learning methods and multi-wavelength photometric data from Chandra, HST, and archival catalogs. The pipeline is designed to be applicable across galaxies with different environments, enabling population studies on a much larger scale than previously possible.

Key references: Yang et al. 2022 (ApJ, 941, 104) — CSC 2.0 ML classification; Chen et al. 2023 (ApJ, 948, 59) — NGC 3532 testbed; Marentes & Rangelov 2025 (RNAAS, 9, 30) — M31 X-ray sources; Rangelov et al. 2024 (ApJ, 961, 12) — Chandra observations of 13 Fermi sources.

Multi-wavelength observations of X-ray source populations

I use multi-wavelength data from Chandra, HST, XMM-Newton, and other observatories to characterize X-ray source populations and their environments. A major focus has been the spatial relationship between X-ray binaries and star clusters in nearby galaxies. In the Antennae merger, a high fraction (~20–25%) of XRBs reside within their parent clusters, and chance superposition accounts for only 1–2% of the observed coincidences. More recently, we have extended this work to M31 using the Panchromatic Hubble Andromeda Treasury, finding that XRBs are significantly more concentrated within 100 pc of clusters than a randomly distributed population.

I also discovered the first pulsar bow shocks in the far-ultraviolet around millisecond pulsars PSR J0437–4715 and PSR J2124–3358 using HST.

Key references: Rangelov et al. 2011 (ApJ, 741, 86) — NGC 4449; Rangelov et al. 2012 (ApJ, 758, 99) — Antennae; Rangelov et al. 2016 (ApJ, 831, 129) and 2017 (ApJ, 835, 264) — FUV bow shocks; Guerrero & Rangelov 2026 (RNAAS, 10, 93) — M31 spatial correlations.

Binary stellar evolution and population synthesis

To connect the observed X-ray source populations with theoretical predictions, I am developing a large-scale binary stellar evolution population synthesis pipeline. Using the binary_c rapid evolution code, I simulate the formation and evolution of X-ray binaries, double compact objects, and gravitational-wave merger progenitors across a range of initial conditions. The goal is to compare synthetic populations directly with the observed ones — testing whether current binary evolution models can reproduce the spatial distributions, luminosity functions, and demographics seen in real galaxies.

This work is ongoing.


Other directions

Pillar II

Space Exploration

Leading the Space Lab at Texas State — an experimental program where students design CubeSat payloads, build balloon-borne instruments, and work with vacuum systems for space-qualified hardware.

CubeSats Balloon missions Student lab
Visit Space Lab site
Pillar III

AI-Driven Discovery

Developing autonomous research agents that combine large language models with domain expertise to accelerate scientific hypothesis generation and literature synthesis.

AutoResearch LLM agents