Flatiron Institute Center for Computational Astrophysics Pre-Doctoral Program

Julianne Dalcanton (Director)

• I am primarily interested in astrophysics on sub-100pc scales, as traced by the ISM and individual stars. My work is primarily observational, grounded in very large data sets, but is highly motivated by, and closely coupled to, theory.
• Project #1: “Spectrospatial forward modeling of the neutral ISM at high resolution”: The recently completed Local Group L-Band Survey offers unprecedented views of neutral hydrogen in the nearest star-forming galaxies. With both high signal-to-noise and exquisite velocity resolution, these new data reveal the tremendous complexity of the ISM, which unfortunately requires new analysis tools to characterize. Newly developed techniques to forward model the joint distribution of position and velocity substructures offer a promising way forward to measuring the detailed cloud-scale properties of the ISM. The resulting measurements have close connections with the physics of turbulence and star formation feedback. This project would be jointly mentored by Julianne Dalcanton (co-PI of the LGBLS) and Thomas Hilder (incoming postdoc, and developer of the “spectracles” package for forward modeling of IFU data).
• Project #2: “Mining large massive star catalogs”: My collaborators and I currently have vast quantities of UV-optical HST observations across the faces of almost all star forming Local Group galaxies. These data allow many different opportunities for exploring the massive star populations, including their evolutionary pathways and connection to the ISM. There are a variety of possible projects that could emerge from this data set, depending on mutual interest. There are also many other CCA collaborators that can contribute to theoretical aspects of potential projects.

Adrian Price-Whelan (Associate Research Scientist, CCA)

• I am broadly interested in galactic dynamics, stellar surveys, and data analysis methods and make heavy use of data from Gaia, SDSS-V, and other stellar spectroscopic surveys. I am interested in projects that intersect any of those things, including related to stellar streams (observationally or theoretically), wide binary star systems, kinetic theory, the planet-star-galaxy connection, data-driven stellar spectroscopy, and more. Right now, I am particularly excited about Gaia Data Release 4 (epoch astrometry) and the Roman space telescope.

Carolina Cuesta Lazaro (Associate Research Scientist, CCA + NYU)

• Project #1: “Data-driven noise modeling for Gravitational Waves detection and inference with generative models” (Co-mentored with: Konstantin Leyde, Max Isi, Francisco Villaescusa-Navarro): Gravitational wave parameter estimation typically relies on the assumption that detector noise is Gaussian, but in reality the noise in detectors like LIGO and Virgo is far more complex — it is non-stationary, non-Gaussian, and shaped by a wide range of instrumental and environmental effects. The goal of this project is to use generative machine learning models to learn the actual noise distribution directly from real data, conditioned on auxiliary witness channels that monitor these disturbances. A key part of the work will be understanding which neural network architectures are most effective at capturing this complex conditional distribution, conditioned on the auxiliary noise channels. Once we have a reliable learned noise model, we plan to use it in two ways. First, for detection: replacing the current likelihood with a data-driven one informed by witnesses could improve sensitivity to signals that are currently missed because they overlap with non-Gaussian noise features, and could potentially improve both template and template-free searches that go beyond standard matched filtering. Second, for parameter estimation: combining the learned noise model with differentiable waveform models would allow us to do inference without the biases that come from misspecifying the noise, whether through likelihood-based or simulation-based approaches. In the longer term, the learned noise models could also serve as environments for training reinforcement learning agents that actively correct instrumental effects in real time, following the spirit of recent work using RL to optimize detector control systems (Buchli et al. 2025).
• Project #2: “Machine Learning for Cosmology” (Co-mentored with: Francisco Villaescusa-Navarro): The upcoming data releases from DESI and the Simons Observatory will map the large-scale structure and the cosmic microwave background at unprecedented scale, but fully exploiting these datasets is far from straightforward. The density field is non-Gaussian, systematics are complex and survey-dependent, and much of the richest cosmological information lives in cross-correlations between different tracers and wavelengths — correlations that are difficult to model with traditional two-point approaches. Machine learning makes it possible to move beyond summary statistics toward field-level inference, working directly with the observed maps to capture non-linear information. The challenge, though, is robustness: models trained on simulations must perform reliably on real observations despite the inevitable mismatch between the two. This project will revolve around ML methods for joint, field-level analysis of cosmological data and their cross-correlations, with an emphasis on building techniques that are robust to the gap between our simulations and the real universe.

Francisco Villaescusa-Navarro (Research Scientist, CCA) with Jonah Rose (FRF, CCA)

• (Francisco Villaescusa-Navarro): I am interested in exploring and quantifying AI agents’ capabilities to push the boundaries of science.
• (Jonah Rose): My research lies at the intersection of computational astrophysics, cosmology, and machine learning, with a primary focus on disentangling the effects of baryonic feedback from fundamental dark matter physics. I utilize the DREAMS simulations, a collection of thousands of cosmological hydrodynamical simulations, to simultaneously explore the impact of diverse astrophysical feedback models and cosmological parameters on galactic structure. I am particularly interested in applying novel machine learning techniques to this dataset to robustly quantify theoretical uncertainties and extract cosmological constraints from small-scale structures. I invite pre-doctoral researchers to propose projects that leverage the statistical power of DREAMS to investigate topics ranging from the satellite galaxy populations and near-field cosmology to the development of robust field-level inference pipelines for next-generation surveys.
• Project #1: “AI agents for scientific discovery”: Recent advances in AI agents are beginning to transform how scientific research is conducted. AI agents are enabling automation of complex tasks and finding new ways to tackle difficult problems.Our group has recently developed Denario (https://github.com/AstroPilot-AI/Denario), a multi-agent research system capable of generating hypotheses, searching the literature, designing research plans, writing and executing code, creating figures, and drafting scientific manuscripts. Our team comprises researchers from diverse fields, including astrophysics, biology, chemistry, materials science, mathematics, machine learning, medicine, neuroscience, planetary science, quantum physics and even philosophy.
  
  In this project, interns will work with the Denario team to explore how AI agents can accelerate scientific discovery. Possible research directions include:
  – Designing and improving agentic architectures for complex research workflows;
  – Developing rigorous benchmarks to evaluate the scientific capabilities of AI agents;
  – Testing agent capabilities across different scientific domains;
  – Applying reinforcement learning to improve agent performance.

  We welcome students with backgrounds in:
  – Machine learning, AI, or computer science;
  – Any scientific discipline (no astrophysics background required).

  This project is ideal for students interested in building next-generation AI systems that interact deeply with real scientific problems. Experience with LLMs or AI agents is a plus, but not required. The mentor for this project is Francisco Villaescusa-Navarro

• Project #2: “Non-Parametric Reconstruction of the Matter Power Spectrum”: Current constraints on the nature of dark matter often rely on specific, parametric models (e.g. Warm Dark Matter or sterile neutrinos) that dictate the shape of the power spectrum suppression. This project aims to relax these assumptions by running a new suite of N-body zoom-in simulations of Milky Way-like halos that vary the initial matter power spectrum non-parametrically, allowing for a model-independent measurement of the true power spectrum shape from Milky Way satellite abundances. A key challenge in this regime is “mode mixing”, where non-linear structure formation entangles the initial linear modes, making it difficult to recover the primordial suppression signal at late times. As mode-mixing is highly non-linear, a machine learning approach will be key to recovering the initial suppression. This project will focus on designing a new suite of simulations, built for a machine learning analysis, that the student will use to reconstruct the initial linear power spectrum from present-day satellite counts and properties. The mentors for this project are Jonah Rose and Francisco Villaescusa-Navarro

Hayk Hakobyan (Flatiron Software Research Fellow, CCA)

• Project #1: “Radiative & quantum electrodynamic processes near compact objects”: [Background] Surroundings of neutron stars and black holes are arguably the most extreme “laboratories” where we can study the otherwise unachievable states of matter in the most exotic regimes. While the physics of energy extraction, particle acceleration, and the dynamics of large-scale flows are controlled by the interaction of magnetic fields with the hot plasma, in many cases (arguably, the majority) radiation becomes a key player, coupling to charged particles via quantum-electrodynamic (QED) interactions (such as Compton scattering, single- and multi-photon pair-production/-annihilation). These phenomena — only recently becoming available for particle-based plasma simulations — are largely ignored in fluid-based frameworks, potentially limiting the scope of interesting problems we can study using (magneto-)hydrodynamic (MHD) simulations. In particular, in most neutron stars QED interactions are the main source of the existing plasma in the system, while in black hole accretion disks these processes are thought to control the production of luminous high-energy flares and sustain the emission of hot regions — so-called coronae. [Goals] The goal of this project would be to explore the dynamics of turbulent plasmas in regime where radiation drag and the dynamics of photons become important using particle simulations, and develop a subgrid model to encompass the effects of radiation translatable to the existing (radiative-)MHD frameworks. [Skills acquired]: Co-developing and running particle- and fluid-based plasma codes (particle-in-cell and radiative-MHD algorithms; mostly C++ & post-processing in Python); Plasma physics, basics of radiative processes & quantum-electrodynamics, thermodynamics of photon-dominated plasmas; General analytic skills on quantifying diffusion/advection/heating in plasmas.

Mike Grudić (Associate Research Scientist-Software, CCA):

• I am interested in developing software and techniques for running and analyzing detailed multi-physics simulations of astrophysical systems, with a focus upon open problems in galaxy and star formation, particularly those involving stellar feedback processes. My main tools are the GIZMO mesh-free multi-physics code and the STARFORGE framework for 3D MHD simulations of ISM dynamics and star cluster formation.
• Project #1: “Bridging spatial scales in numerical simulations of star formation”: In collaboration with Dr. Shivan Khullar, the student would explore methodology and parameter space for numerical simulations of star-forming giant molecular clouds with the GIZMO code, with the overall aim of a self-consistent physical numerical solution connecting the scales of galaxies and individual stars. Two possible areas of focus are: (1) Numerical methodology for adaptive hyper-Lagrangian resolution refinement, exploring refinement criteria and techniques for resolving key scales in star-forming molecular clouds while minimizing numerical noise and ensuring convergence. (2) Star formation as a boundary condition: developing a novel numerical setup wherein star formation is driven by continuous inflow from larger scales, rather than the in-situ collapse of a pre-existing dense cloud. The project would explore the implications of this scenario for global evolution, star formation, star cluster properties, and stellar demographics, vis-a-vis traditional in-situ models.
• Project #2: “Connecting the physics of HII regions with observables”: In collaboration with Dr. Lachlan Lancaster, the student would use or develop semi-analytic and/or 3D MHD models of HII region evolution, with the aim of establishing the connection between the physical properties of HII regions with observables across the electromagnetic spectrum accessible to current and future telescopes. Possible areas of focus include: (1) Modeling of nebular emission from 3D MHD simulations of HII regions using the COLT radiative transfer code. This project would involve running a parameter study of HII regions using a pre-existing numerical framework, and post-processing the outputs at various phases of evolution. Many simplifying assumptions are traditionally made when inferring physical properties from nebular lines, and these will be examined or revised in light of the realistic 3D models. (2) Post-processing a novel semi-analytic HII region model developed at the CCA to produce synthetic observations connecting with various existing observations. Depending on the student’s interests, this could include observables in X-ray (tracing hot gas), nebular emission from photoionized gas, or fine-structure emission (e.g. CII [158μm]) from neutral gas.

Rachel Somerville (Senior Research Scientist, CCA)

• Galaxy formation, co-evolution of galaxies and SMBH, early galaxies and BH
• Project #1: “Supermassive black hole mergers and their multimessenger signatures”: (co-mentored alongside Kunyang (Lily) Li): We are currently working on improving the modeling of black hole seeding, accretion, and feedback in the Santa Cruz Semi-analytic model, motivated especially by the exciting recent discoveries of surprisingly massive black holes at very early times with the James Webb Space Telescope. One big missing piece is proper modeling of how black holes merge within galaxies and how frequently they may be ejected from their galactic hosts by dynamical effects like gravitational recoil and slingshot. Black hole mergers are thought to be one of the main channels for BH growth in the very early Universe, so this could have a big impact on how many black holes can be seen with JWST, as well as of course impacting the number of gravitational wave events that may someday be detected with LISA. There are also interesting tensions between current model predictions and the claimed detection of the stochastic gravitational wave background recently reported by Pulsar Timing Arrays. We plan to implement more detailed modeling of BH mergers and dynamics into the Santa Cruz SAM using the RAMCOAL sub-grid modeling approach developed by Li and collaborators (2025, A&A, 701, 232). RAMCOAL is also being implemented within cosmological simulations with the numerical hydrodynamic code RAMSES to create the Coscoal simulations. We invite project proposals that would exploit these tools to make predictions for science questions related to merging of galaxies and SMBH, and their observability via electromagnetic or gravitational wave signatures, in the early or late Universe.
• Project #2: “Modeling of nebular line emission in cosmological simulations” (co-mentored alongside Yurina Nakazato, and L. Y. Aaron Yung): Line emission from ionized HII regions (nebular line emission) is an important tracer of large scale structure in large volume galaxy surveys (such as those that will be carried out by the Nancy Grace Roman Space Telescope or the ongoing SPHEREx mission), and also provides probes of galaxy physical properties such as gas density, temperature, and metallicity. We have developed an approach to model nebular emission in the Santa Cruz SAMs, using an approach similar to the one described in Hirschmann et al. (2023, MNRAS, 526, 3610). We are interested in using these models to make predictions for nebular emission line luminosity functions and line ratios from redshift 0 to ~10, and in comparing our predictions with observations from SDSS, JWST, and other facilities. We are also interested in using predictions from the high resolution cosmological FirstLight simulations (Ceverino et al. 2017; 2026; see also Nakazato project description) to validate and refine the ‘subgrid’ treatment of nebular emission in SAMs, with a focus on high redshift galaxies.

Thomas Pfeil (Flatiron Research Fellow, CCA)

• Protoplanetary disks, hydrodynamics, radiation-hydrodynamics, turbulence, planet-disk interactions, planetesimal formation.
• Project #1: “Radiation, Hydrodynamics and Dust Evolution in Protoplanetary Disks” (co-mentored alongside Phil Armitage): The planet formation group at CCA, in collaboration with scientists at the University of Munich and MPIA, has developed the TriPoD dust coagulation model for 3D hydrodynamic simulations of protoplanetary disks with Athena++. We welcome project proposals utilizing this newly developed tool for computational studies of protoplanetary disks with dust evolution, e.g.: Studies of the thermal structure of protoplanetary disks via radiation-hydrodynamic simulations; Simulations of substructure formation via hydrodynamic instabilities; Planetesimal formation in substructures; Planet-disk interactions.
• Project #2: Applications of the 1D dust evolution code TriPoDPy (co-mentored alongside Phil Armitage): The one-dimensional version of the TriPoD dust coagulation model, TriPoDPy, allows for the very fast computation of realistic dust size distributions in protoplanetary disks. This makes population studies and the fitting of complex protoplanetary disk models to observational data possible. Possible projects could include: Probabilistic inference of dust properties from observational data; Disk population synthesis studies; Planetesimal formation in substructures.

Yan-Fei Jiang (Research Scientist, CCA)

• I am generally interested in numerical simulations of accretion disks around different kinds of systems, including compact objects and planets, particularly related to the effects of thermodynamics on disk structures. Examples include in structures of AGNs, migration in protoplanetary disks, effects of dust on disk dynamics. I am also interested in stellar structures, particularly massive stars, cosmic ray and radiation feedback in galaxies, as well as numerical simulations of transient systems such as TDEs and QPEs. I am exploring new numerical algorithms for simulations.

Yurina Nakazato (Flatiron Research Fellow, CCA)

• High-redshift galaxies (before/during the Epoch of Reionization, z ≳ 6); Zoom-in cosmological simulations; Emission line/dust attenuation & re-emission calculations; Interpretation of the latest ALMA/JWST observations
• Project #1: “Star-formation fluctuation” (co-mentored alongside Rachel Somerville): Recent JWST observations have revealed high-redshift galaxies at z ≳ 6 with diverse star-formation histories. For example, JWST has observed (mini-)quenched galaxies at z > 7 (Looser+24, Weibel+25), with the most recent candidate at z = 11.04 (Harikane+26). At the same time, bursty star formation has been suggested as a mechanism to alleviate the tension between theoretical predictions and the unexpectedly high number density of z>10 galaxies detected by JWST. Using the FirstLight cosmological zoom-in simulations (Ceverino+17, 26), we plan to investigate star-formation fluctuations (mainly focusing on mini-quenching) statistically by utilizing a large number of snapshots and calculating number densities. We also plan to calculate emission lines (Hα, UV continuum) and SEDs (Balmer break) to directly compare with JWST observations. This project will also identify galaxies in highly bursty phases, which may exhibit nebular-continuum-dominated SEDs.
• Project #2: “Dust Properties and MIRI Predictions for High-Redshift Galaxies” (co-mentored alongside Rachel Somerville): Recent JWST and ALMA observations have revealed the dust attenuation and re-emission properties of high-redshift galaxies at z ≳ 6. JWST can measure dust attenuation properties such as the UV slope (β), attenuation (A_V), and attenuation curves, while ALMA can measure re-emission properties such as L_IR, dust emissivity (β_d), and dust temperatures. We have over 4000 snapshots for z = 6–9 galaxies in the FirstLight simulations with post-processed dust radiative transfer outputs, enabling statistical investigations of these properties. We plan to directly compare them with JWST/ALMA observational results and study the underlying physical mechanisms. IIn JWST Cycles 3 and 4, several follow-up MIRI observations (imaging and spectroscopy) targeting high-redshift galaxies have been conducted, with results forthcoming. Combining MIRI imaging with NIRCam results is expected to improve the precision of SED-fitting-based estimates of physical properties (e.g., stellar mass). With over 4000 snapshots of SEDs and MIRI/NIRCam mock imaging from the FirstLight zoom-in simulations, we can study how adding MIRI observations improves the accuracy of SED-fitting results. Prospective students can choose to focus on either project: global dust properties or MIRI predictions. Potential mentors are Rachel Somerville and Yurina Nakazato. Since Yurina Nakazato has expertise in analyzing FirstLight simulation outputs and access to the data, we propose the above projects using FirstLight. We also have access to other zoom-in simulations using RAMSES for z > 10 galaxies. Prospective students are welcome to propose projects investigating similar or related science questions using these or other simulations they have access to.

Elena Pinetti (Flatiron Research Fellow, CCA)

• I am interested in multi-messenger searches for dark matter in novel cosmic environments, including cosmic voids, filaments, and blank-sky fields. These underexplored regions provide reduced astrophysical backgrounds and therefore offer particularly clean settings in which to test particle dark matter scenarios across a broad mass range. My work combines theoretical modeling with direct comparison to observational data from both space- and ground-based facilities. I use existing datasets from instruments such as the James Webb Space Telescope to search for faint spectral signatures and diffuse emission consistent with dark matter annihilation or decay. In parallel, I develop forecasts for next-generation observatories—including CTAO, SKA, and Euclid. The goal of my research is to identify signatures of dark matter while discriminating them from conventional astrophysical sources.
• Project 1: “Dark matter searches with X-ray data”: Sub-GeV dark matter (DM) candidates—including MeV-scale particles and keV-scale decaying DM—can produce observable X-ray emission through both prompt and secondary channels. Annihilation or decay into electron–positron pairs generically leads to inverse Compton (IC) scattering of ambient photon fields, generating diffuse X-ray emission. In parallel, keV-scale DM can produce monochromatic or narrow spectral features in the soft X-ray band, offering a distinctive spectroscopic signature. Recent theoretical studies have shown that X-ray observations provide some of the most stringent constraints on sub-GeV DM parameter space. This project aims to search for dark matter–induced signals in archival Chandra blank-sky data. By exploiting blank-sky fields—where conventional astrophysical contamination is minimized—the project seeks to probe sub-GeV dark matter with X-ray observations.
• Project 2: “Dark matter searches with cosmic voids”: Cosmic voids occupy most of the volume of the Universe and provide a uniquely clean laboratory for studying dark matter and galaxy formation in low-density environments. Because voids are highly underdense, the processes that regulate galaxy growth—such as gas accretion, feedback, and mergers—can differ significantly from those operating in filaments and clusters. As a result, the luminosity function and halo occupation statistics of void galaxies offer a sensitive probe of both baryonic physics and the underlying dark matter distribution.Hydrodynamical simulations such as IllustrisTNG provide an interesting framework in which to study these effects. The aim of this project is to derive the luminosity functions of galaxies residing in cosmic voids using state-of-the-art hydrodynamical simulations. The student will identify voids within large simulation volumes and construct the luminosity function of galaxies in voids. These results will be incorporated into cross-correlation analyses between large-scale structure tracers and diffuse radiation fields (e.g., X-ray or γ-ray backgrounds), with the goal of disentangling standard astrophysical contributions from potential dark matter signals. By improving the modeling of galaxy populations in underdense regions, this work will enhance the robustness of indirect dark matter searches that rely on the clean environments of cosmic voids.