About

Julie Comerford is a Professor in the Department of Astrophysical and Planetary Sciences at the University of Colorado, Boulder. Her research focuses on galaxy mergers, active galactic nuclei (AGN), and gravitational waves. She studies galaxy and supermassive black hole properties and their implications for the gravitational waves produced by binary supermassive black holes detectable by pulsar timing arrays and LISA. Her work also involves using dual AGN and offset AGN to trace AGN feeding in galaxy mergers, as well as examining AGN outflows to understand feedback effects between an AGN and its host galaxy. Her research group includes graduate students and postdoctoral researchers working on observations, simulations, and the astrophysics of black holes and galaxy interactions. She has contributed to the understanding of supermassive black hole binaries, galaxy merger processes, and the gravitational wave background, and her work has been featured in various media outlets and scientific publications.

Research topics

• Computer Science
• Astronomy
• Astrophysics
• Artificial Intelligence
• Physics
• Astrobiology

Selected publications

• Beyond the Brightest: A Deep Learning Approach to Identifying Major and Minor Galaxy Mergers in CANDELS at z ∼ 1

  Zenodo (CERN European Organization for Nuclear Research) · 2026-04-15

  datasetOpen access

  Abstract: Galaxy mergers play an important role in many aspects of galaxy evolution, therefore, more accurate merger identifications are paramount for achieving a complete understanding of galaxy evolution. As we enter the era of large, deep, high-resolution imaging surveys, we are able to observe mergers extending to even lower masses and higher redshifts. Despite low-mass galaxies being more common, many previous merger identification methods were mostly calibrated for high-mass galaxies which are easier to identify. To prepare for upcoming surveys, we train a convolutional neural network (CNN) using mock $HST$ CANDELS images at $z\sim1$ created from the IllustrisTNG50 cosmological simulation. We successfully identify galaxy mergers between a wide range of galaxies ($10^8M_\odot < M_\star < 10^{12.5}M_\odot$, and $q\geq1:10$), achieving overall accuracy, purity, and completeness of $\sim65\%$. We show, for the first time, that a CNN trained on this diverse set of galaxies is capable of identifying major mergers, especially at early stages ($74\%$ accuracy), similar to that of networks trained at lower redshifts and/or higher masses (with accuracies ranging between $66-80\%$). We discuss the inherent limits of galaxy merger identification due to orientation angle, finding $98\%$ of mergers are correctly identified from at least one angle, and $61\%$ from the majority of angles. We additionally explore the confounding variables, such as star formation, to consider when applying to real data. This network enables the exploration of the impact of previously overlooked mergers of high mass ratio and low stellar masses on galaxy evolution in CANDELS, and can be expanded to surveys from $JWST$, Rubin, $Roman$, and $Euclid$. Datsets: Dataset directories include train, validation, and test subdirectories, which include both classes within subdirectories. Code to produce these datasets is found on Github. All datasets were split into an 70%/15%/15%train/validation/test split. Merger Tables: The catalog of galaxies from TNG50 that we use to create our images, along with merger stage, star formation rate, and stellar mass information. z1mocks: The mock image dataset in physical units (erg/cm^2/s/$\mathring{A}$ ). z1mocks_norm: The mock image dataset normalized with asinh stretch, ready to input to the CNN. z1mocks_1filtgrey: The mock image dataset normalized with asinh stretch, in only the F160W filter, ready to input to the CNN. z1mocks_3filtgrey: The mock image dataset normalized with asinh stretch, with all three filters summed into one input channel, ready to input to the CNN. nano_1_trial1_lr3.154957151204336e-07_epoch120.pth: The saved weights from our model at the epoch used for the test set. These weights were used to produce all plots in the paper.

  PublisherDOI

• Beyond the Brightest: A Deep Learning Approach to Identifying Major and Minor Galaxy Mergers in CANDELS at z ∼ 1

  Zenodo (CERN European Organization for Nuclear Research) · 2026-04-15

  datasetOpen access

  Abstract: Galaxy mergers play an important role in many aspects of galaxy evolution, therefore, more accurate merger identifications are paramount for achieving a complete understanding of galaxy evolution. As we enter the era of large, deep, high-resolution imaging surveys, we are able to observe mergers extending to even lower masses and higher redshifts. Despite low-mass galaxies being more common, many previous merger identification methods were mostly calibrated for high-mass galaxies which are easier to identify. To prepare for upcoming surveys, we train a convolutional neural network (CNN) using mock $HST$ CANDELS images at $z\sim1$ created from the IllustrisTNG50 cosmological simulation. We successfully identify galaxy mergers between a wide range of galaxies ($10^8M_\odot < M_\star < 10^{12.5}M_\odot$, and $q\geq1:10$), achieving overall accuracy, purity, and completeness of $\sim65\%$. We show, for the first time, that a CNN trained on this diverse set of galaxies is capable of identifying major mergers, especially at early stages ($74\%$ accuracy), similar to that of networks trained at lower redshifts and/or higher masses (with accuracies ranging between $66-80\%$). We discuss the inherent limits of galaxy merger identification due to orientation angle, finding $98\%$ of mergers are correctly identified from at least one angle, and $61\%$ from the majority of angles. We additionally explore the confounding variables, such as star formation, to consider when applying to real data. This network enables the exploration of the impact of previously overlooked mergers of high mass ratio and low stellar masses on galaxy evolution in CANDELS, and can be expanded to surveys from $JWST$, Rubin, $Roman$, and $Euclid$. Datsets: Dataset directories include train, validation, and test subdirectories, which include both classes within subdirectories. Code to produce these datasets is found on Github. All datasets were split into an 70%/15%/15%train/validation/test split. Merger Tables: The catalog of galaxies from TNG50 that we use to create our images, along with merger stage, star formation rate, and stellar mass information. z1mocks: The mock image dataset in physical units (erg/cm^2/s/$\mathring{A}$ ). z1mocks_norm: The mock image dataset normalized with asinh stretch, ready to input to the CNN. z1mocks_1filtgrey: The mock image dataset normalized with asinh stretch, in only the F160W filter, ready to input to the CNN. z1mocks_3filtgrey: The mock image dataset normalized with asinh stretch, with all three filters summed into one input channel, ready to input to the CNN. nano_1_trial1_lr3.154957151204336e-07_epoch120.pth: The saved weights from our model at the epoch used for the test set. These weights were used to produce all plots in the paper.

  PublisherDOI

• Mapping AGN winds: A connection between radio-mode AGNs and the AGN feedback cycle

  Astronomy and Astrophysics · 2024 · 8 citations

  • Physics
  • Astrophysics
  • Astronomy

  We present a kinematic analysis based on the large integral field spectroscopy (IFS) dataset of SDSS-IV MaNGA (Sloan Digital Sky Survey/Mapping Nearby Galaxies at Apache Point Observatory; ∼10 000 galaxies). We have compiled a diverse sample of 594 unique active galactic nuclei (AGNs), identified through a variety of independent selection techniques, encompassing radio (1.4 GHz) observations, optical emission-line diagnostics (BPT), broad Balmer emission lines, mid-infrared colors, and hard X-ray emission. We investigated how ionized gas kinematics behave in these different AGN populations through stacked radial profiles of the [O III] 5007 emission-line width across each AGN population. We contrasted AGN populations against each other (and non-AGN galaxies) by matching samples by stellar mass, [O III] 5007 luminosity, morphology, and redshift. We find similar kinematics between AGNs selected by BPT diagnostics compared to broad-line-selected AGNs. We also identify a population of non-AGNs with similar radial profiles as AGNs, indicative of the presence of remnant outflows (or fossil outflows) of past AGN activity. We find that purely radio-selected AGNs display enhanced ionized gas line widths across all radii. This suggests that our radio-selection technique is sensitive to a population in which AGN-driven kinematic perturbations have been active for longer durations (potentially due to recurrent activity) than in purely optically selected AGNs. This connection between radio activity and extended ionized gas outflow signatures is consistent with recent evidence that suggests radio emission (expected to be diffuse) originated due to shocks from outflows. We conclude that different selection techniques can trace different AGN populations not only in terms of energetics but also in terms of AGN evolutionary stages. Our results are important in the context of the AGN duty cycle and highlight integral field unit data’s potential to deepen our knowledge of AGNs and galaxy evolution.

  PublisherOA PDFDOI

• The Molecular Gas in the NGC 6240 Merging Galaxy System at the Highest Spatial Resolution

  American Astronomical Society Meeting Abstracts #235 · 2020

  • Computer Science
  • Artificial Intelligence
  • Astronomy

  We present the highest resolution --- 15 pc (0.03'') --- ALMA $^{12}$CO(2-1) line emission and 1.3mm continuum maps, tracers of the molecular gas and dust, respectively, in the nearby merging galaxy system NGC 6240, that hosts two supermassive black holes growing simultaneously. These observations provide an excellent spatial match to existing Hubble optical and near-infrared observations of this system. A significant molecular gas mass, $\sim$9$\times$10$^9$M$_\odot$, is located in between the two nuclei, forming a clumpy stream kinematically dominated by turbulence, rather than a smooth rotating disk as previously assumed from lower resolution data. Evidence for rotation is seen in the gas surrounding the southern nucleus, but not in the northern one. Dynamical shells can be seen, likely associated with nuclear supernovae remnants. We further detect the presence of significant high velocity outflows, some of them reaching velocities $&gt;$500 km/s, affecting a significant fraction, $\sim$11\% of the molecular gas in the nuclear region. Inside the spheres of influence of the northern and southern supermassive black holes we find molecular masses of 7.4$\times$10$^8$M$_\odot$ and 3.3$\times$10$^9$M$_\odot$, respectively. We are thus directly imaging the reservoir of gas that can accrete onto each supermassive black hole. These new ALMA maps highlight the critical need for high resolution observations of molecular gas in order to understand the feeding of supermassive black holes and its connection to galaxy evolution in the context of a major galaxy merger.

• The Gravitational View of Massive Black Hole Mergers

  Bulletin of the American Astronomical Society · 2019-05-31 · 6 citations

  article

  Coalescing, massive black-hole (MBH) binaries are the most powerful sources of gravitational waves (GWs) in the Universe, which makes MBH science a prime focus for ongoing and upcoming GW observatories. The Laser Interferometer Space Antenna (LISA) – a gigameter scale space-based GW observatory – will grant us access to an immense cosmological volume, revealing MBHs merging when the first cosmic structures assembled in the Dark Ages. LISA will unveil the yet unknown origin of the first quasars, and detect the teeming population of MBHs of 104−7 M⊙ forming within protogalactic halos. The Pulsar Timing Array, a galactic-scale GW survey, can access the largest MBHs the Universe, detecting the cosmic GW foreground from inspiraling MBH binaries of ∼109 M⊙. LISA can measure MBH spins and masses with precision far exceeding that from electromagnetic (EM) probes, and together, both GW observatories will provide the first full census of binary MBHs, and their orbital dynamics, across cosmic time. Detecting the loud gravitational signal of these MBH binaries will also trigger alerts for EM counterpart searches, from decades (PTAs) to hours (LISA) prior to the final merger. By witnessing both the GW and EM signals of MBH mergers, precious information will be gathered about the rich and complex environment in the aftermath of a galaxy collision. The unique GW characterization of MBHs will shed light on the deep link between MBHs of 104−1010 M⊙ and the grand design of galaxy assembly, as well as on the complex dynamics that drive MBHs to coalescence.

  Publisher

• Multi-Messenger Astrophysics with Pulsar Timing Arrays

  arXiv (Cornell University) · 2019-03-18 · 26 citations

  preprintOpen access

  Pulsar timing arrays (PTAs) are on the verge of detecting low-frequency gravitational waves (GWs) from supermassive black hole binaries (SMBHBs). With continued observations of a large sample of millisecond pulsars, PTAs will reach this major milestone within the next decade. Already, SMBHB candidates are being identified by electromagnetic surveys in ever-increasing numbers; upcoming surveys will enhance our ability to detect and verify candidates, and will be instrumental in identifying the host galaxies of GW sources. Multi-messenger (GW and electromagnetic) observations of SMBHBs will revolutionize our understanding of the co-evolution of SMBHs with their host galaxies, the dynamical interactions between binaries and their galactic environments, and the fundamental physics of accretion. Multi-messenger observations can also make SMBHBs 'standard sirens' for cosmological distance measurements out to $z\simeq0.5$. LIGO has already ushered in breakthrough insights in our knowledge of black holes. The multi-messenger detection of SMBHBs with PTAs will be a breakthrough in the years $2020-2030$ and beyond, and prepare us for LISA to help complete our views of black hole demographics and evolution at higher redshifts.

  PublisherOA PDFDOI

• Multiwavelength Studies of Dual AGN in the Swift/BAT Sample

  AAS · 2018-01-01

  article
  Publisher

• Dissecting the Butterfly: Dual Outflows in the Dual AGN NGC 6240

  American Astronomical Society Meeting Abstracts #231 · 2018-01-01

  article
  Publisher

• Supermassive Black Hole Binaries: Multi-Messenger Astrophysics and Long Baselines with the Next-Generation Very Large Array

  American Astronomical Society Meeting Abstracts #231 · 2018-01-01

  article
  Publisher

• Offset Active Galactic Nuclei

  2018-12-01

  article
  Publisher

Frequent coauthors

• Sarah Burke-Spolaor

  4 shared

• George C. Privon

  National Radio Astronomy Observatory

  3 shared

• Jenny E. Greene

  3 shared

• Ezequiel Treister

  3 shared

• Tamara Bogdanović

  3 shared

• Laura Blecha

  3 shared

• Alberto Sesana

  2 shared

• C. M. Urry

  2 shared

Labs

• Julie Comerford LabPI