arXiv subjects

Rafael Martínez-Galarza

Publications and source records attributed to Rafael Martínez-Galarza.

5 recordsLinked to original sources

arXiv2026

Speeding up Gravitational Lens Mass Models with Machine Learning: Applications in X-ray Astronomy

Multi-wavelength observations of quadruply lensed quasars constitute a powerful probe of cosmology, dark matter substructure along the line of sight, and the structure of X-ray emitting regions in high-redshift quasars. These investigations are conditional on acquiring an accurate model for the surface mass density of matter lensing these quasars. We propose a simulation-based machine learning method to accelerate parameter inference in real quadruply lensed systems by several orders of magnitude. We simulate a grid of quadruply lensed sources with Singular Isothermal Ellipsoid (SIE) lenses and use the projected positions of the four lensed images to train two fully connected neural networks that predict the mass parameter and ellipticity. For a large fraction of simulated systems, the neural network-initialised mass models converge in time-scales of a few minutes and recover the source position at the <0.''005 level for a broad range of lens masses and ellipticities. We apply our neural networks to seven quadruply lensed quasars, lensed by isolated galaxies or a galaxy-perturber pair, which have archival Chandra observations. The final optimised mass models for each quasar predict the observed lensed image positions in Gaia Data Release 3. These mass models enable the caustic method, which locates the X-ray-to-optical emission regions to milliarcsecond precision in these otherwise unresolvable systems, improving the effective angular resolution of Chandra at high-z by up to two orders of magnitude. Our approach accelerates this mass modelling by supplying informed initial parameters, enabling application to the many new quadruply lensed systems expected from forthcoming surveys.

astro-ph.GA↗

arXiv2026

Augmenting representations with scientific papers

Astronomers have acquired vast repositories of multimodal data, including images, spectra, and time series, complemented by decades of literature that analyzes astrophysical sources. Still, these data sources are rarely systematically integrated. This work introduces a contrastive learning framework designed to align X-ray spectra with domain knowledge extracted from scientific literature, facilitating the development of shared multimodal representations. Establishing this connection is inherently complex, as scientific texts encompass a broader and more diverse physical context than spectra. We propose a contrastive pipeline that achieves a 20% Recall@1% when retrieving texts from spectra, proving that a meaningful alignment between these modalities is not only possible but capable of accelerating the interpretation of rare or poorly understood sources. Furthermore, the resulting shared latent space effectively encodes physically significant information. By fusing spectral and textual data, we improve the estimation of 20 physical variables by 16-18% over unimodal spectral baselines. Our results indicate that a Mixture of Experts (MoE) strategy, which leverages both unimodal and shared representations, yields superior performance. Finally, outlier analysis within the multimodal latent space identifies high-priority targets for follow-up investigation, including a candidate pulsating ULX (PULX) and a gravitational lens system. Importantly, this framework can be extended to other scientific domains where aligning observational data with existing literature is possible.

cs.LG↗

arXiv2025

Encoding and Understanding Astrophysical Information in Large Language Model-Generated Summaries

Large Language Models have demonstrated the ability to generalize well at many levels across domains, modalities, and even shown in-context learning capabilities. This enables research questions regarding how they can be used to encode physical information that is usually only available from scientific measurements, and loosely encoded in textual descriptions. Using astrophysics as a test bed, we investigate if LLM embeddings can codify physical summary statistics that are obtained from scientific measurements through two main questions: 1) Does prompting play a role on how those quantities are codified by the LLM? and 2) What aspects of language are most important in encoding the physics represented by the measurement? We investigate this using sparse autoencoders that extract interpretable features from the text.

cs.CL↗

arXiv2023

The Future of Astronomical Data Infrastructure: Meeting Report

The astronomical community is grappling with the increasing volume and complexity of data produced by modern telescopes, due to difficulties in reducing, accessing, analyzing, and combining archives of data. To address this challenge, we propose the establishment of a coordinating body, an "entity," with the specific mission of enhancing the interoperability, archiving, distribution, and production of both astronomical data and software. This report is the culmination of a workshop held in February 2023 on the Future of Astronomical Data Infrastructure. Attended by 70 scientists and software professionals from ground-based and space-based missions and archives spanning the entire spectrum of astronomical research, the group deliberated on the prevailing state of software and data infrastructure in astronomy, identified pressing issues, and explored potential solutions. In this report, we describe the ecosystem of astronomical data, its existing flaws, and the many gaps, duplication, inconsistencies, barriers to access, drags on productivity, missed opportunities, and risks to the long-term integrity of essential data sets. We also highlight the successes and failures in a set of deep dives into several different illustrative components of the ecosystem, included as an appendix.

astro-ph.IM↗

arXiv2023

Disentangling the AGN and Star-Formation Contributions to the Radio-X-ray Emission of Radio-Loud Quasars at 1<z<2

We constrain the emission mechanisms responsible for the prodigious electromagnetic output generated by active galactic nuclei (AGN) and their host galaxies with a novel state-of-the-art AGN radio- to-X-ray spectral energy distribution model fitting code (ARXSED). ARXSED combines multiple components to fit the spectral energy distributions (SEDs) of AGN and their host galaxies. Emission components include radio structures such as lobes and jets, infrared emission from the AGN torus, visible-to-X-ray emission from the accretion disk, and radio-to-ultraviolet emission from the host galaxy. Applying ARXSED to the radio SEDs of 20 3CRR quasars at 1 < z < 2 verifies the need for more than a simple power law when compact radio structures are present. The non-thermal emission contributes 91%-57% of the observed-frame 1.25mm to 850μm flux, and this component must be accounted for when using these wavelengths to estimate star-formation properties. We predict the presence of strong radio-linked X-ray emission in more than half the sample sources. ARXSED estimates median (and the associated first and third quartile ranges) BH mass of $2.9_{1.7}^{6.0} \times 10^9~\rm M_{\odot}$, logarithm of Eddington ratio of $ -1.0_{-1.2}^{-0.6} $, and spin of $ 0.98_{0.94}^{0.99} $ for our sample. The inferred AGN torus and accretion disk parameters agree with those estimated from spectroscopic analyses of similar samples in the literature. We present the median intrinsic SED of the luminous radio-loud quasars at 1 < z < 2 ; this SED represents a significant improvement in the way each component is modeled.

astro-ph.HE↗