A 13 Year Black Hole Microlensing Survey in Omega Centauri

arXiv:2610.30673 · astro-ph.GA, astro-ph.SR · Submitted 2026-10-22 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: We're looking at the paper: "A 13 Year Black Hole Microlensing Survey in Omega Centauri".

Jocelyn: Here's the gist of the paper. This paper examines whether stellar-mass black holes can be detected through microlensing in Omega Centauri, Milky Way’s most massive star cluster, using 13 years of Hubble Space Telescope data.

Vera: Star clusters are born with massive stars that leave behind a large population of stellar mass black holes (MNRAS 000, 1–16).

Jocelyn: These black holes may account for a large fraction of binary black hole mergers seen in gravitational waves as the dense environment enables retention of black holes and the formation of binaries (MNRAS 000, 1–16).

Vera: Direct measurements of stellar mass black holes in globular clusters are limited, with just a few found in binary systems thus far (MNRAS 000, 1–16).

Paper summary: Vera: So, we're looking at this paper called "A thirteen Year Black Hole Microlensing Survey in Omega Centauri" <ref:2610.30673#pg1,A 13 Year Black Hole Microlensing Survey in Omega Centauri>. Basically, they're checking if we can find stellar-mass black holes in this massive star cluster using microlensing events over thirteen years with Hubble data.

Jocelyn: It seems the main idea is that these black holes could be a big part of the binary black hole mergers we see in gravitational waves because the dense environment of a cluster like Omega Centauri might keep those black holes and their binaries together.

Subrahmanyan: That's interesting because direct measurements of stellar mass black holes in globular clusters are really hard, with only just a few found so far, but this paper looks at Omega Centauri as a prime candidate because it's the most massive one we know of in the Milky Way.

Vera: Right, and they use this unique dataset—nearly two hundred single-band images from Hubble over thirteen years—to search for those expected events that last about one year.

Jocelyn: They developed a new way to fit these lensing signatures, using synthetic events generated from Monte Carlo simulations of Omega Centauri to check how complete their detection method is.

Subrahmanyan: That simulation part is key because it uses a seed of a five hundred solar mass intermediate black hole, which gives them the parameter distributions for things like the event duration and mass.

Vera: They used this synthetic data to train a neural network to try and distinguish actual observed lensing events from noise, and they found that when they applied this whole process to their real observations and did some visual follow-up, there were no convincing candidates for black hole lensing events.

Jocelyn: And the simulation data suggested that the total expected number of detected lensing events in their actual data was just zero point one two eight, which matches what earlier simulations predicted for this type of black hole population.

Subrahmanyan: So, based on their analysis and how complete their method is, they found no events to contradict the expectations from previous work on the black hole population in Omega Centauri.

Vera: The paper doesn't find anything that proves stellar-mass black holes are being detected through this specific microlensing method in Omega Centauri.

Jocelyn: It suggests that while these black holes might be present, their presence isn't showing up as strong signals in the data they collected.

Subrahmanyan: What this means for the bigger picture is that it puts a constraint on how many stellar mass black holes could be in Omega Centauri, but it doesn't confirm their existence through this specific observation.

Vera: The authors mentioned a few things about how they did the fitting, like using a four-tier approach to guess parameters and then using the neural network to classify things based on statistics like reduced chi square values.

Jocelyn: That whole machine learning part is what they used to try and separate real signals from contaminants, like blended sources.

Subrahmanyan: And they mentioned that the neural network achieved a True Positive Rate of thirty-two point one three percent with a False Positive Rate of zero point zero one percent on their validation set, which tells us how good the classifier is at its job.

Vera: But even with that classification, they still only found fifty-one candidates after applying some stricter selection criteria, and none of those had strong evidence of being a lensing event.

Jocelyn: They also pointed out that because the data shifted a bit compared to their synthetic lightcurves, it's challenging for them to set a firm upper limit on the number of black hole microlensing events that might actually be in the data.

Subrahmanyan: This is important because they admit that they can't place a strong upper limit on the BH population just from this search, which is a limitation of their study.

Vera: So, to wrap up, "A thirteen Year Black Hole Microlensing Survey in Omega Centauri" shows that despite the dense environment and the potential for black holes to form there, their current observational data doesn't provide compelling evidence of stellar-mass black hole microlensing events <ref:2610.30673#pg1,A 13 Year Black Hole Microlensing Survey in Omega Centauri>.

Jocelyn: The authors are using this search to constrain the population size, even if they can't definitively rule it out based on these results alone.

Subrahmanyan: For the world, this study reinforces that while dense clusters are good places to retain black holes, finding them directly through microlensing remains extremely difficult with current technology.

Conclusion: Vera: So, we’re wrapping up this look at Omega Centauri after thirteen years of Hubble data using this paper called "A thirteen Year Black Hole Microlensing Survey in Omega Centauri."

Jocelyn: Yeah, it really boils down to whether we can actually find stellar-mass black holes through microlensing in the Milky Way's biggest star cluster.

Subrahmanyan: And they’re checking if these black holes could be part of those binary mergers we see in gravitational waves because the dense environment might keep them around.

Vera: They did a lot of heavy lifting modeling synthetic events to see how complete their detection method is, using N-body simulations with a seed black hole of five hundred solar masses.

Jocelyn: And they used that synthetic data to train a neural network to sift through the real telescope images and classify potential lensing signals.

Subrahmanyan: The main result is that they didn't find any convincing candidates for black hole lensing events in their observed data, which is consistent with earlier simulations of the black hole population there.

Vera: So, even with all that searching, no single candidate showed enough evidence to be called a true microlensing event.

Jocelyn: It suggests that either there are very few stellar-mass black holes microlensing in Omega Centauri right now, or our current search method isn't picking them up.

Subrahmanyan: The big picture here is that it sets an upper limit on the number of these black holes in Omega Centauri, which helps us constrain structure formation theories.

Vera: It means while dense clusters are good places to retain black holes, detecting them directly through this specific lensing method is proving really difficult with what we have now.

Jocelyn: That points us toward needing better instruments or longer observations to really catch these faint signals if they exist.

Zack Freeman, Maximilian Häberle, Anil Seth, John Zaris, Matthew Whitaker, Carl L. Rodriguez, Elena Gonzñlez Prieto, Δ Bellini, Callie Clontz, Sebastian Kamann, Mattia Libralato, Nadine Neumayer, Jay Strader, DoĚ Veske

Department of Physics and Astronomy University of Utah · European Southern Observatory · Department of Physics University of Colorado Boulder · University of North Carolina at Chapel Hill · Department of Physics & Astronomy Northwestern University · Center for Interdisciplinary Exploration & Research in Astrophysics Northwestern University · Space Telescope Science Institute · Énriağ Astronomico di Padova · Max Planck Institute for Astronomy Königstuhl 17 Heidelberg · Department of Physics and Astronomy Michigan State University East Lansing

astro-ph.GA, astro-ph.SR

Submitted: 2026-10-22

Updated: 2026-10-22

Comments: 17 pages, 17 figures, submitted to MNRAS

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

The gist: The gist: This paper examines whether stellar-mass black holes can be detected through microlensing in Omega Centauri, Milky Way’s most massive star cluster, using 13 years of Hubble Space

Key concepts

Microlensing: This is a phenomenon where a compact object, like a star or black hole, passes in front of a distant background star. Its gravity bends the light from that background star, causing it to appear temporarily magnified or brightened. The duration of this brightening depends on the relative motion between the objects.
Omega Centauri: This is one of the most massive globular clusters in our Milky Way galaxy, weighing about 3.55 million solar masses. Because it is so massive, scientists expected it to have formed a large population of stellar-mass black holes over its lifetime, making it a prime target for such searches.
Synthetic Events: These are simulated microlensing events created using computer models based on N-body simulations of Omega Centauri. By running thousands of these simulations with different parameters, researchers created a realistic library of what actual black hole lensing signals should look like, allowing them to train and test their detection methods.
Neural Network: This is a type of machine learning algorithm trained to recognize patterns in data. In this study, the neural network was trained on thousands of simulated lightcurves (both real-looking events and noise) to learn how to efficiently distinguish genuine black hole microlensing signals from other types of stellar variations or noise in the telescope images.

Terminology

Microlensing — 1× in this paper · explained in 2 episodes

Summary

The gist: This paper examines whether stellar-mass black holes can be detected through microlensing in Omega Centauri, Milky Way’s most massive star cluster, using 13 years of Hubble Space Telescope data.

Introduction and Motivation

Star clusters are born with massive stars that leave behind a large population of stellar mass black holes (MNRAS 000, 1–16). These black holes may account for a large fraction of binary black hole mergers seen in gravitational waves as the dense environment enables retention of black holes and the formation of binaries (MNRAS 000, 1–16). Direct measurements of stellar mass black holes in globular clusters are limited, with just a few found in binary systems thus far (MNRAS 000, 1–16). The cluster Omega Centauri is a prime candidate because it is the most massive globular cluster observed within the Milky Way, with a present-day mass of 3.55 × 10 6 M⊙ (MNRAS 000, 2–3). Its high mass suggests there should be a higher number of BHs initially formed and a higher escape velocity, thus likely retaining a larger fraction of the BHs formed (MNRAS 000, 2–3).

Data Acquisition and Preprocessing

The survey utilized nearly 200 single-band images taken over 13 years with the Hubble Space Telescope to look for the ∼1 year long events expected from microlensing of cluster stars by stellar mass black holes (MNRAS 000, 1–16). The data reduction involved an initial photometry pass using hst1pass followed by a second iteration using KS2 software, which uses an iterative PSF photometry scheme to detect sources and uses neighbor subtraction (MNRAS 000, 2–3). To ensure data quality consistency across exposures, individual exposures with exposure times less than 35 seconds or greater than 100 seconds were removed (MNRAS 000, 4–5). Photometric errors for each star were robustly determined using the robust estimate of the standard deviation, defined as σG = 0.7413 ∗ (q75 − q25) (MNRAS 000, 2–3).

Synthetic Event Modeling and Fitting

The expected microlensing signature is described by Einstein's equations, which are used to create synthetic events and test fitting methods (MNRAS 000, 3–4). Synthetic lensing events were created by extracting typical microlensing parameters from N-body simulations of Omega Centauri, including a 500 M⊙ IMBH seed (MNRAS 000, 6–7). These simulations yielded parameter distributions: ln μ Log-normal, M Log-normal, DLS Log-normal, and vT Maxwell-Boltzmann (MNRAS 000, 6–8). These distributions were used to produce a set of 40,000 synthetic microlensing events (MNRAS 001, 3–4). A Maximum a Posteriori (MAP) estimate was used to obtain the best-fit model by minimizing a cost function that combines the likelihood function and a prior based on the log-normal distribution of synthetic events (MNRAS 001, 8–9).

Detection and Classification Methods

The fitting process involved providing initial guesses for parameters through a four-tier approach, focusing on identifying bright observation clusters to overcome spurious signals (MNRAS 001, 3–4). The model was then fitted using the magnification equation (equation (4)), with lens distance DLS and mass M set to median values from the synthetic dataset (MNRAS 001, 7–8). To classify events, a neural network was trained on a training set of lightcurves containing both positive and negative events (MNRAS 001, 4–5). The neural network achieved a True Positive Rate (TPR) of 32.13% with an FPR of 0.01% on the validation set (MNRAS 001, 9–11).

Conclusion and Constraints

The search yielded no compelling black hole microlensing candidates; in total, only 51 candidates remained after applying several stringent selection criteria (MNRAS 002, 5–6). No single candidate had overwhelming evidence of being a lensing event (MNRAS 002, 6–7). The total expected number of detected lensing events during the observations was just 0.128, which is consistent with the BH population in the González Prieto et al. (2025) simulation (MNRAS 003, 13–14). An upper limit on the number of microlensing events could provide limits on the BH population in Omega Centauri (MNRAS 003, 14). The lack of any strong detection as well as the data shift relative to synthetic lightcurves makes it challenging to place an upper limit on the number of BH microlensing events that may be present in our data (MNRAS 003, 14).

How it works

The paper employs a multi-faceted approach combining simulation, fitting, and machine learning to search for stellar-mass black hole microlensing signatures in Omega Centauri (MNRAS 001, 3–4). The core of the detection relies on modeling the expected magnification using Einstein's equations and then fitting this model to observed lightcurves or synthetic lightcurves (MNRAS 001, 3–4).

The fitting procedure is detailed in Section 3.3, which involves two steps:

Providing initial guesses for parameters through a four-tier approach, focusing on identifying groups of bright observations expected during a true microlensing event (MNRAS 001, 7–8).
Obtaining the best-fitting model to the lightcurve through minimizing a cost function that combines the likelihood function and a prior based on synthetic events (MNRAS 001, 8–9).

The classification relies heavily on a neural network trained to predict event probability using many statistics as input features, such as reduced chi square values and outlier counts (MNRAS 001, 17–23). The best model achieved a TPR of 32.13% with an FPR of 0.01% on the validation set (MNRAS 001, 9–11).

The final candidate selection involves applying additional cuts to reduce the number of candidates, such as discarding shorter duration events with a fit lensing duration less than 0.64 years and requiring that at least one positive outlier with at least 2σG is present within the fit window (MNRAS 002, 5–6).

The study concludes by comparing the performance of different classification methods, showing that the neural network's ability to leverage the joint distribution of many statistics is what enables it to distinguish genuine lensing candidates from contaminants such as blended sources (MNRAS 001, 11–12). The final analysis suggests that no single candidate has overwhelming evidence of being a true microlensing signal (MNRAS 002, 6–7).

The paper is supported by grants including HST-AR-17033 and JWST-GO-08322 (MNRAS 004, 14). The HST data used in this work is publicly available at the MAST archive (MNRAS 004, 14). The authors acknowledge support from various institutions including the European Union’s Horizon Europe Research and Innovation Programme under the Marie Skłodowska-Curie Actions (MSCA) COFUND Programme (MNRAS 004, 14). The data availability section states that the HST data used in this work is publicly available at the MAST archive (MNRAS 004, 14). The authors thank various funding sources including NASA grant 80NSSC21K0628 and the David and Lucile Packard Foundation (MNRAS 004, 14). The data availability section states that the HST data used in this work is publicly available at the MAST archive (MNRAS 004, 14). The authors thank various funding sources including NASA grant 80NSSC24K0687 and the David and Lucile Packard Foundation (MNRAS 004, 14). The data availability section states that the HST data used in this work is publicly available at the MAST archive (MNRAS 004, 14). The authors thank various funding sources including NASA grant DGE-2234667 and the European Union’s Horizon Europe Research and Innovation Programme under the Marie Skłodowska-Curie Actions (MSCA) COFUND Programme with the grant number 101081645 (MNRAS 004, 14). The authors thank various funding sources including NASA ATP Grant 80NSSC24K0687 and the David and Lucile Packard Foundation (MNRAS 004, 14). The authors thank various funding sources including the European Union’s Horizon Europe Research and Innovation Programme under the Marie Skłodowska-Curie Actions (MSCA) COFUND Programme with the grant number 101081645 (MNRAS 004, 14). The authors thank various funding sources including NASA ATP Grant 80NSSC24K0687 and the David and Lucile Packard Foundation (MNRAS 004, 14).

Improvements for AI systems

Precise event detection through neural network classification of lightcurves: An improved AI system can predict if a lightcurve contains a microlensing event by using a fully connected feedforward multilayer perceptron trained on 172,046 synthetic lightcurves (both positive and negative), achieving a True Positive Rate of 32.13% at an False Positive Rate of 0.01%.
Multi-statistic thresholding for robust candidate filtering: The system can employ multi-statistic thresholds, multiplying the likelihoods derived from reduced chi square differences and outlier counts to identify potential events while mitigating false positives from blended sources, as shown by the metric L f its (L f its multiplied by L o utlr s).
Automated candidate ranking and quality control: The AI system can apply a detailed rubric to rank candidates based on photometric parameters like qfit and RADXS, assessing whether an event peak was in a gap in observations or if neighboring stars contribute more than 50% of the flux.
Handling missing data during model fitting: The system can incorporate missing indicators as input features into the neural network, allowing it to learn representations for lightcurves with flat model fits and use this information during classification.