Flatiron Institute Center for Computational Astrophysics Pre-Doctoral Program

Mentor Projects

Below is a list of CCA staff who are interested in mentoring predoctoral researchers. Some of the projects listed are general interests, some specific project ideas, and some contain both. For projects with multiple mentors, the primary mentor is listed first. If you want to work with one of the mentors listed below on a project related to their interests that differs from the specific project(s) they listed, feel free to reach out and pitch it.

Please note that full-time CCA senior staff without joint faculty positions are given priority due to their limited access to students via other channels. These individuals are listed first, and potential mentors with joint faculty positions are listed afterward, as indicated below.

For reference, here is a guide to the Flatiron titles listed below:

FRF = Flatiron Research Fellow (postdoc; will co-mentor with a more senior scientist)
ARS = Associate Research Scientist (assistant professor level)
RS = Research Scientist (associate professor level)
SRS = Senior Research Scientist (professor level)

Julianne Dalcanton (Director)

I am interested in the intersection of data and theory, with a particular focus on “data at high resolution in the Local Universe” — basically, scales where we can resolve astrophysics on the scale of 10s of parsecs. I am particularly interested in the connections between stars, stellar evolution, and the interstellar medium.

Jared Goldberg (FRF)

Research interests: My interests broadly lie at the intersection of stellar structure and evolution, and massive star outcomes as core-collapse supernovae and the resulting transient emission. I am also interested in other variable stellar activity such as stellar pulsations and low-frequency variability. To explore these objects and phenomena, I make use of 1D stellar evolution/explosion modeling (using MESA + SNEC or STELLA) and 3D radiation-hydrodynamics simulations (with Athena++), and recently have developed an excitement to collaborate with galactic modelers to explore the effects of these processes in their galactic contexts. I am happy to be of broader support and/or to work together to create other potential projects, and I’m open to collaboration across CCA!

Investigating the impact of single-star-resolution supernova feedback in low-mass galaxies [with Ulrich Steinwandel (FRF), Matteo Cantiello (RS) and Rachel Somerville (SRS)]

With advances in numerical techniques and computing power, simulations of low-mass (dwarf) galaxies can now resolve down to single-star precision, enabling direct explorations of the impact of stellar physics and its uncertainties on the properties of the star-forming Interstellar Medium (ISM). With this project, we will build upon the work of Steinwandel & Goldberg (2023) which used the meshless-finite-mass code pgadget-3 to explore the impact of variable supernova explosion energy and successful/failed explosion outcomes on galactic/ISM properties (e.g. star formation rate, mass and energy-loading, and the phase structure of the ISM). Building on this framework, the student will run and analyze a denser grid of models varying the supernova explosion energy, successful supernova explosion fraction, and (novelly!) the metallicity in order to explicitly derive and numerically test scaling relationships between the properties of the populations of supernova explosions and galactic/ISM properties.

Coupling pulsations and 3D convection to unify early supernova emission with red supergiant circumstellar material [with Matteo Cantiello (RS) and Yan-Fei Jiang (ARS)]

Many supernova observations, especially of the recent nearby core-collapse Supernova 2023ixf in M101, have re-ignited the field’s excitement about circumstellar material (CSM) around Red Supergiant stars. SN2023ixf, for example, was observed with a strongly pulsating progenitor star exploding just months after the radial maximum of its 1000-day pulsation period, with ample material observed around the star across different wavelengths and inferred from both supernova observations and pre-explosion imaging. The big question: How can you get that much stuff around the star (without an outburst)? Using the open-source 1D stellar evolution and pulsation codes MESA+GYRE, the pre-doctoral fellow will model the (radially pulsating) progenitor of this event, and make it explode at the appropriate pulsation phase. The student will also use information about the 3D convective inhomogeneities motivated by 3D RHD simulation data, coupled with these pulsating stellar models, to create a theoretically-motivated profile of the circumstellar material around the star. These explosion models will then be compared to the observations in order to test the hypothesis of the pulsational origin of the inferred circumstellar material.

Ore Gottlieb (FRF) with Matteo Cantiello (RS) and Lorenzo Sironi (RS)

Research interests: multi-messenger astrophysics, black holes, black hole-neutron star mergers, neutron star-neutron star mergers, supernovae, gamma-ray bursts, fast blue optical transients, binary systems, accretion disks, active galactic nuclei, relativity, magnetohydrodynamics, radiation processes, neutrino & cosmic-ray emission, non-coherent gravitational waves, numerical simulations

Connecting stellar death and jet formation: simulating the dynamo effect in collapsing stars

The collapse of some rapidly rotating massive stars may initiate the formation of long gamma-ray bursts (LGRBs) – the most luminous explosions in the Universe. LGRBs are powered by relativistic magnetized jets emerging from the black hole formed in the core of the stellar progenitor. The development of these Poynting-flux dominated jets necessitates the presence of strong poloidal magnetic fields threading the black hole. However, the differential rotation in the star predominantly generates magnetic fields of a toroidal geometry. This underscores the significance of a dynamo effect that converts the toroidal field into a poloidal one for jet formation. This project, utilizing advanced 3D general-relativistic magnetohydrodynamic simulations, aims to numerically investigate the dynamo effect in collapsing massive stars and decipher how jets are born. This will establish a self-consistent connection between the demise of massive stars and the formation of jets for the first time.

Connecting jet precession and observational signatures: modeling non-thermal emission from wobbling jets

Relativistic jets are widespread in the universe, observed across various astrophysical systems such as binary compact mergers, collapsing stars, and the centers of galaxies. Traditionally, jets have been presumed to be axisymmetric, leading to the modeling of their emission based on this assumption. However, recent advancements in state-of-the-art simulations have unveiled the inherent wobbling and precession of jets, posing a challenge to existing jet emission models. This revelation emphasizes the pressing need to compute the electromagnetic emission signatures of these oscillating jets, compare the results with current observations, and explore their underlying physics. This project employs data derived from cutting-edge simulations of wobbling gamma-ray burst (GRB) jets formed subsequent to the merger of two compact objects, and potentially involves conducting additional simulations, focusing on their non-thermal (prompt and afterglow) emission. The nature of the project can be tailored to be either more numerical or semi-analytic, contingent upon the student’s preferences and aptitudes.

Chris Hayward (RS)

Research interests: I work on a variety of topics in galaxy formation. I mainly use hydrodynamical simulations, typically those from the Feedback in Realistic Environments (FIRE) project, of which I am a PI. I also employ analytic toy models, semi-analytic and semi-empirical models, and occasionally observations (usually via collaborations, but I’ve led some observational work myself). Much of my work involves performing radiative transfer to generate synthetic observations to more directly compare simulations and observations (i.e., forward-modeling).

Specific current interests include: star formation self-regulation via stellar feedback and the relation to bursty star formation and outflows; infrared-selected galaxies; dust; and galaxy protoclusters. Previous pre-docs whom I co-mentored have worked on the effect of cosmic rays on thermal instability, developing a neural network-based emulator for radiative transfer, hierarchical modeling of galaxies’ dust attenuation curves, and how well infrared-luminous galaxies trace protoclusters. I’m open to discussing possible projects related to any of the above interests, but especially the following topics:

  • AGN feedback (with incoming FRF Samuel Ward)
  • Studying the origin of the first black holes at high redshift with cosmological simulations (with incoming FRF Christian Partmann)
  • dusty star-forming galaxies (DSFGs)
  • JWST synthetic observations

Lucia Perez (FRF) with Shy Genel (RS)

Research interests: cosmological inference, galaxy formation, galaxy surveys, cosmological simulations, galaxy spectra, galaxy photometry, structure formation, semi-analytic modeling

While many advanced methods for the statistical analysis of cosmological data have been developed to handle the looming breadth of new astronomical observations, a common constraint is the availability of training cosmological simulation data, especially that which includes realistic galaxy formation physics and the volume and/or resolution necessary to match observations. CAMELS-SAM and its newest updates offer crucial and unique training data sets of realistic galaxies across an enormous range of cosmologies and galaxy physics formulations. As the larger-volume ‘hump’ of the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project, CAMELS-SAM uses semi-analytic models (SAMs) of galaxy formation to flexibly and quickly generate galaxies over 1000 dark-matter only simulations of L=100 h^-1 cMpc and N=640^3 covering a broad cosmological parameter space in OmegaM and sigma8.

There are several expansions to CAMELS-SAM that will be available for a pre-doctoral student to analyze. In particular:

  • new galaxy catalogs using the Santa Cruz SAM with 9 total varied astrophysical parameters;
  • updated galaxy catalogs using the Santa Cruz SAM with SEDs and complete photometry generated for all galaxies (likely for both the 3-astro parameter and 10-astro parameter catalog groups, TBD);
  • and new galaxy catalogs with the wholly separate L-Galaxies SAM.

These CAMELS-SAM datasets together will offer many possibilities for the next generation of simulation-based inference for galaxy formation and cosmology. In particular, the expansion of astrophysical parameters can address questions of model misspecification; leveraging L-Galaxies could yield models more robust to galaxy formation prescriptions; and the SC-SAM catalogs with SEDs and photometry open up possibilities for machine learning with direct observables.

We solicit proposals that will lean into the core of CAMELS–Cosmology and Astrophysics with MachinE Learning–explicitly with these new CAMELS-SAM expansions. We also welcome proposals to create new data using the CAMELS-SAM infrastructure, such as incorporating other SAMs or methods of populating dark matter halos with galaxies, or applying post-processing pipelines to generate observables interesting to both students and mentors.

Adrian Price-Whelan (ARS)

Research interests: Dark matter, galactic dynamics, Milky Way structure and evolution, data analysis and statistical inference, binary stars, methods, community software

Finding Dark Matter Subhalo Impacts in Stellar Streams

Stellar streams form from disrupting stellar systems, like dwarf galaxies and globular clusters. Streams from star clusters are dynamically “cold” and are therefore very sensitive to perturbations from passing massive objects, like dark matter subhalos. These perturbations leave imprints in the density and velocity structure of streams. We now know of >100 streams around the Milky Way. We will model the membership and density structure of a few of the most prominent streams using a flexible statistical modeling framework for modeling the stream and any background stellar density with data from the Gaia mission. We will have to build in support for handling the Gaia selection function to obtain robust measurements. This work will be done with the “Community Atlas of Tidal Streams” (CATS) collaboration.

Searching for Stellar Streams

Stellar streams form from disrupting stellar systems, like dwarf galaxies and globular clusters. Streams from star clusters are dynamically “cold” and are therefore very sensitive to perturbations from passing massive objects like dark matter subhalos, which make them extremely useful objects to find within the Milky Way. We will develop a new method to search for streams using data from the Gaia mission.

Rachel Somerville (SRS) with Viraj Pandya, Aklant Bhowmick and Aaron Yung

Research interests: first galaxies, co-evolution of supermassive black holes and galaxies, galaxy clustering

Modeling the first black holes

The James Webb Space Telescope has discovered a surprisingly large number of accreting supermassive black holes (Active Galactic Nuclei; AGN) at very early times in the Universe. There are hints that these black holes are overmassive relative to their galaxies — i.e., that they lie above the scaling relations for SMBH and galaxies observed at lower redshift — yet this may be a selection effect. In this project, we will use the new, ultra-fast JAX-based semi-analytic model Sapphire (Pandya et al. 2023) to explore different models for black hole seeding and accretion and their implications for black hole accretion rate distributions, black hole-galaxy scaling relations, and AGN bolometric luminosity functions at ultra-high redshifts (z>5). The black hole seeding models will be guided by the results of high resolution hydrodynamic simulations recently carried out by Bhowmick et al. (2023, 2024).

Depending on the interests of the student and the available time, there are several different aspects of the project that could also be explored. 1) We could exploit an Implicit Likelihood Simulation Based Inference approach (e.g. Ho et al. 2024) to estimate the multi-dimensional posteriors of the parameters relating to BH growth using available observations of AGN and BH scaling relations. 2) We could couple the semi-analytic models with spectral synthesis models to predict multi-wavelength observable properties of high-z AGN, including emission lines. 3) We could explore predictions for black hole merger rates and gravitational wave emission, relevant for the upcoming LISA mission and Pulsar Timing Arrays.

What can we learn from galaxy clustering in deep narrow surveys?

The James Webb Space Telescope excels at extremely deep observations, but will never be able to survey very large areas. In addition, many surveys will have sparse observations of accurate spectroscopic redshifts, with most galaxies having only more imprecise photometric redshifts. In this project, we will use mock JWST lightcones created using semi-analytic models to explore what constraints are possible on galaxy models, the galaxy-halo connection, and potentially cosmology, from feasible future JWST observations of galaxy clustering. Depending on the students’ expertise and interests, we could utilize machine-learning based methods or a more traditional approach.

Laura Sommovigo (FRF) with Rachel Somerville (SRS) and Chris Hayward (RS)

Research interests: dust, ism, galaxy formation, high-redshift Universe

Missing accreting black holes in the high-redshift Universe

Over its first year of operation, the James Webb Space Telescope has broken multiple records, providing us with rest-frame Ultraviolet detections of the most ancient galaxies and supermassive black holes (SMBH) discovered to date. However, the bulk of the Active Galactic Nuclei (AGN) population is likely composed of low-luminosity and dust-obscured systems, which are still completely undetected beyond redshift z≈5. Investigating the missing AGN population at high redshift is paramount to addressing the open question of the mechanisms responsible for the formation and surprisingly efficient growth of supermassive black holes in the early Universe. Over the last decade, large surveys at lower redshift (z=0-2) have allowed us to gather large statistical samples of thousands of AGN, with detailed multiwavelength observations. Despite this rapid observational advancement, in terms of physical modeling, we are somewhat stuck to the unified AGN model from the 90s. According to this model, most of the UV absorption and reprocessing in AGN hosts is powered by the dust located in a ring-like dense substructure, referred to as the AGN torus. A consensus on the modeling of the torus and its quantitative effect on the spectral energy distribution (SED) of AGNs is yet to be reached. Relying on the new comprehensive multiwavelength AGN datasets, we intend to apply Bayesian model selection between the existing models for AGN spectral energy distributions, to find the best fitting one(s). We will then apply such AGN SED model to cosmological simulations developed in-house at CCA to obtain new accurate constraints on the AGN luminosity function at high-z, quantifying the current observational bias in our understanding of the accreting BH population in the early Universe.

Primordial galaxies: a glimpse of a polluted future?

Shortly after the first data from JWST were released, the detection of “too many, too bright, too massive” galaxies as early as redshift z>10 (a few hundred million years after the Big Bang) created a frenzy in the high-z galaxy community. Statements such as “JWST detects galaxies breaking the ΛCDM paradigm” made it into the public press. Almost a year later, the dust has settled, and we know now that the tension between JWST observations and theoretical models is likely explained by small (i.e. sub-galactic) scale physics rather than broken cosmology. Nevertheless, there is an elephant in the room that has mostly been ignored so far: the effect of dust obscuration. Albeit young, these JWST-detected galaxies at z>10 are extraordinarily compact and efficiently star-forming. Thus, dust production will skyrocket in such systems, strongly impacting their visibility and inferred properties from spectral energy fitting, possibly re-opening the Pandora’s box of the discrepancy with theoretical predictions. Building on local-universe physical dust models, we aim to investigate the impact of the extreme interstellar medium conditions characterizing these starbursting systems on the microscopic properties of dust grains (e.g. grain size distribution, chemical composition, etc.). We will further investigate the large-scale (galactic to cosmological) implications of accounting for dust physics in these controversial systems.

Lieke van Son (FRF) with Matteo Cantiello (RS)

Research interests: gravitational-wave sources, massive stars, binary stars, stellar mass black holes, neutron stars

Can we create the 35 Msun bump with chemically homogeneously evolving stars?

Gravitational-wave observations have revealed features in the mass distribution of merging binary black holes (BBH). The feature that has received by far the most attention is the bump at 35 Msun in the distribution of more massive BBH components. For the last 5 years, this feature has virtually always been attributed to a pile-up resulting from ‘pulsational pair-instability supernova’ (PPISN). However, recently it has become increasingly clear that the location of this feature is in tension with the expected location of pair-instability supernovae. Whether or not this feature is from PPISNe links to fundamental questions like “What is the most massive BH we can form from stars?” and “What merging BBHs are formed from isolated binaries vs in dynamical environments?”

If the bump at 35 Msun is not from PPISN, then what is it? In this project, we will explore one of the few channels that is in the running: chemically homogeneously evolving stars. Chemically homogeneously evolving stars are stars that have no classic ‘core-envelope’ structure. Instead, meridional circulations due to rapid rotation cause the stars to mix throughout and evolve homogeneously. Homogeneous evolution is more easily achieved for more massive stars, naturally pushing the peak in their mass distribution to ~30 Msun. Yet much about their evolution, especially how abundant such stars are, remains uncertain. We will investigate what mass distribution we expect for chemically homogeneous stars given existing models, and if it is possible to reproduce the observed peak at 35 Msun within known uncertainties by exploring new models.

Francisco Villaescusa-Navarro (RS)

Research interests: I am interested in combining massive datasets from numerical simulations with machine learning to learn about cosmology, astrophysics, and dark matter. I am particularly interested in geometric deep learning, anomaly detection, and self-supervised learning.

The CAMELS project

The CAMELS project was designed to create bridges between cosmology and galaxy formation by combining numerical simulations with machine learning. CAMELS contains the largest collection of state-of-the-art cosmological hydrodynamic simulations to date, and I am very interested in using machine-learning tools to explore these simulated Universes to learn about the laws and constituents of our own Universe. I am interested in working with students who are interested in contributing to CAMELS from any angle; from students with expertise in running numerical simulations to students interested in analyzing the data. Students with strong (or purely) CS backgrounds are encouraged to apply as no (astro-)physics background is required. More information about CAMELS can be found here and here.

The DREAMS project [with Mariangela Lisanti (ARS)]

DREAMS stands for DaRk mattEr with AI and siMulationS, and it is a project designed to combine astrophysics with particle physics through machine learning to shed light on the nature of dark matter and to learn about the physics that shapes the Universe on small scales. The project will generate the largest set of state-of-the-art hydrodynamic simulations covering different regimes (e.g. Milky-Way, dwarfs…) with different dark matter models and different astrophysics. Machine learning techniques will then be used to identify new observables to unveil the nature of dark matter and to provide tighter constraints on dark matter properties. Our team contains experts in numerical simulations, particle physics, cosmology, and machine learning. We are looking for students interested in any aspect of this project, from running simulations to analyzing them with traditional and/or machine-learning techniques. More information about DREAMS can be found here.

Sebastian Wagner-Carena (FRF) with Francisco Villaescusa-Navarro (RS)

Research interests: I work at the intersection of machine learning and astrophysics research. My interests include dark matter physics, large-scale structure, simulation-based inference, contrastive learning, Bayesian machine learning, strong lensing, and differentiable simulators. I’m open to discussing potential projects relating to any of the above interests.

Sequential Simulation-Based Inference with the CAMELS Simulations

Simulation-based inference (SBI) stands as the most promising tool for extracting the full, small-scale information present in our simulations. Standard SBI methods use a fixed training set to best capture the target probability density. Sequential methods interact directly with the simulation pipeline to generate training examples that are informative for the specific observations in question. The goal of this project would be to leverage the current suite of simulations to demonstrate the power of sequential inference for future extensions of the CAMELS suite.

Hydrodynamics in the Latent Space

This project would focus on building a shared latent space encoding for different hydrodynamic models (i.e. TNG, SIMBA, etc.). The goal of the project would be to build a physically informative mapping in a data-driven fashion that would allow us to 1) express the similarity of competing hydrodynamics prescriptions, 2) directly infer which hydrodynamics prescriptions best align with our universe, and 3) steer our future simulations towards using our observations.

Projects for which the primary senior mentors hold joint faculty positions

Lorenzo Sironi (RS) with Elizabeth Tolman (FRF)

Research interests: high-energy astrophysics, plasma astrophysics, magnetic field generation and dissipation, particle acceleration and non-thermal emission

Tearing-mediated reconnection in magnetohydrodynamic partially-ionized plasmas

Recent work (Tolman, Elizabeth A., et al. “Tearing-mediated reconnection in magnetohydrodynamic poorly ionized plasmas. I. Onset and linear evolution.” arXiv preprint arXiv:2312.14076 (2023)) has described the linear stage of the tearing instability in poorly ionized plasmas. In this project, we will increase the ionization fraction to gain greater understanding of how poorly ionized physics transitions into fully ionized physics. This is of relevance for a large number of partially ionized regions in the Universe (e.g., molecular clouds, proto-planetary disks).

Wenrui Xu (FRF) with Phil Armitage (SRS)

Research interests: protoplanetary disk; planet formation; planetary dynamics; (radiation and/or magneto-) hydrodynamic simulations

Exploring the global picture of circumplanetary disks and giant planet accretion

A sufficiently massive protoplanet can carve out a gap in the protoplanetary disk, and the gas flow around the protoplanet exhibits complex dynamics, including the possibility of forming a circumplanetary accretion disk. Better understanding of the properties of the circumplanetary disk and the accretion onto the protoplanet are crucial for understanding the mass assembly of gas giants and for interpreting recent observations of accreting protoplanets such as PDS-70 b and c. However, the large scale separation between the protoplanet and the disk poses a challenge in obtaining a reliable global picture of the multi-scale flow around the protoplanet through numerical simulations. In this project, we will address this challenge with the help of recent developments in numerical techniques for efficiently simulating multi-scale flows, such as zoom-in, V-cycles, and timestep scaling (don’t worry if you are not familiar with these terms – some are new or unpublished). We will apply these techniques on the hydrodynamic code athena++ and investigate how circumplanetary disk properties and planet accretion rate depend on the angular-momentum transport within the circumplanetary disk and the size of the protoplanet. These results will provide a basic global picture of giant planet accretion. We will also lay the foundation for future studies that incorporate more realistic physical treatments such as radiation transfer.

Advancing Research in Basic Science and MathematicsSubscribe to Flatiron Institute announcements and other foundation updates