This page presents a limited selection of my current and past research activities. See my Google Scholar or ORCID page for an up-to-date list of my publications.

Black holes, broadly

It is believed that certain relativistic jets are powered by the rotational energy of spinning black holes. My research primarily centers on understanding the nature of accretion around such black holes, especially in the context of this jet–hole connection. Although analytic calculation and numerical simulations support the theory, concrete observational evidence of energy extraction remains elusive both because of theoretical uncertainties and the challenge of making detailed observations.

Simulated image of a black hole accretion flow as the inclination angle between the hole and the observer is varied. The cartoon in the bottom right shows the relative inclination between the observer and the disk (parallel to the hole), neglecting the effects of general relativity. As inclination changes, the shape, size, and position of the black hole shadow within the photon ring evolves. The bright wisps correspond to hot features in the accretion flow and jet. Images produced using the ipole code.

The interaction between black holes, event horizon scale plasma physics, and the mechanisms that lead to jet production is complex. In order to understand the interaction and to make detailed predictions about observational signatures of energy extraction, it is first necessary to develop a robust working model of both the accretion physics at horizon scales and the processes that mediate the dynamics of jet launching. My current research focuses on developing methods to overcome theoretical model uncertainties and identify robust features of the accretion flow in order to make definitive observational predictions.

Numerical simulations of black hole accretion flows

To better understand the nature of plasma near supermassive black holes, I use a variety of (radiation) general relativistic magnetohydrodynamics codes to generate numerical fluid simulations of accretion, and I post-process these simulations using polarized radiation transport codes to predict what observations of these accretion flows might look like.

Recently, the Event Horizon Telescope (EHT) collaboration published horizon-scale images of the black hole at the center of the elliptical galaxy M87; I produced numerical models of black hole accretion flows and generated synthetic images that were used to inform the constraints presented in the first EHT sequence papers.

Left panel: reconstructed image of the M87 black hole using observations taken by the EHT in 2017. Center panel: high resolution simulated black hole image. Right panel: the same simulated image in the center panel blurred to the resolution of the EHT data. Color encodes brightness temperature. Figure is reproduced from 10.3847/2041-8213/ab0f43.

The state space that describes possible configurations of the black hole / magnetic field / plasma system is incredibly large and mired by thermodynamic uncertainties (e.g., cooling and the details of dissipation / energy injection mechanisms). Nevertheless, simplified models can be used to constrain various parameters and identify reasonable accretion flow models to make it feasible to perform preliminary numerical surveys.

Snapshot of a numerical simulation of magnetized plasma accreting onto a spinning black hole. The panels (left-to-right) show the density, internal energy, and magnetization of plasma in log scale.

In addition to directly comparing model results to observations (like those from the EHT), it is valuable to compare and contrast models among themselves. I have studied and explored the properties of retrograde accretion (especially in models of magnetically arrested disks), characterized the connection between plasma dynamics in the disk and jet, and identified observational signatures of these physical processes.

Position of Lagrangian "tracer particles" in plasma accreting onto a spinning black hole. The density of tracer particles tracks the density of the plasma, which in this case corresponds to a retrograde accretion flow (angular momentum of hole anti-parallel to accretion flow angular momentum). Rather than maintaining a steady disk of matter all the way to the event horizon, accretion proceeds in filaments.

I have also explored the importance of different physical processes vis-a-vis black hole accretion by helping to identify the regimes in which electron-positron pair plasmas may be relevant and under what conditions bremsstrahlung radiation is non-negligible.

Observables of the Kerr geometry

Decomposition of simulated black hole image showing contributions due to different "photon (sub-)rings". Photon subrings are an observational consequence of strong lensing in the Kerr spacetime near compact objects.

It is possible to sidestep the details of the plasma emission and measure properties of the black hole itself by looking for signatures of general relativity and the Kerr metric, like in the photon ring. In space near a black hole, the effects of gravity are so strong that light may end up on bound orbits, i.e., it is possible that a ray of light will "orbit" around the black hole many times before escaping off to an observer at large distance. The set of bound orbits is called the photon sphere and can produce a characteristic, sharp ring-like feature in images.

The shape and size of the photon ring are directly related to the mass and angular momentum (magnitude and inclination) of the black hole and are independent of the details of the plasma physics in the surrounding accretion flow. By making measurements at long baselines (i.e., making a simultaneous observation of a black hole using telescopes that are very far away from each other), it is possible to directly measure the morphological features of the photon ring.

Other image classification techniques

In addition to taking measurements of the total intensity of light from M87, the EHT also captured information about the polarization of the light, i.e., the orientation of the electromagnetic field as it reached the detectors. When light is emitted due to synchrotron radiation, the polarization of that light is implicitly tied to the orientation of the magnetic field in the source material. The connection between polarization and the magnetic field means that it might be possible further constrain our model of the black hole accretion flow using polarimetric data.

To this end, I have helped develop an analysis framework for differentiating between types of accretion flows according to the degree of azimuthal order in the structure of polarization in black hole accretion flow images.

In addition to using conventional, theory-based methods to connect observables to model parameters, I have also explored the utility of machine learning methods for classifying black hole images. Preliminary results using convolutional neural networks suggest the existence of continuous features in the image domain that map directly to parameters like black hole spin.

Effectiveness of modified ResNet-18 neural network in classifying black hole spin for simulated images drawn from simulations with parameters that were not part of the training set. Color encodes the true value of spin in the simulation, and histogram bins show log number of classification counts. The training image set only contained images of simulations with spins -0.9375, -0.5, 0, 0.5, and 0.9375. It is interesting to note that although the network favored the trained spin values, its predictions were always consistent with nearest-neighbor truth values. Figure reproduced from arXiv:2007.00794.

Code development

I am also a developer of several astrophysics-based codes including:

  • iharm3d — a general relativistic magnetohydrodynamics code,
  • ebhlight — a radiation general relativistic magnetohydrodynamics code,
  • ipole — a polarized radiative transport image-producing code, and
  • igrmonty — a radiative transport spectrum-producing code.

Infectious disease modeling

Since Spring 2020, I have been involved in epidemiological modeling of the spread of the COVID-19 epidemic. I have developed stochastic agent-based simulators and non-Markovian age-of-infection models that are coupled to Markov chain Monte Carlo statistical inference methods. My work has been used to inform policy, and I am a member of advisory committees to the State of Illinois, the local health department, and the University of Illinois system. I have also been a part of research into the relationship between herd immunity and persistent population heterogeneities (e.g., the difference between a delivery person's social contacts versus a stay-at-home computational physicist's ones). The pydemic python package was developed as part of this project.

Comparison of model predictions for hospital (and ICU) bed occupancy and daily deaths against data from the State of Illinois. Uncertainty bounds shrink as more data is incorporated in the calibration procedure. Subsequent model predictions are consistent with previous predictions, even as the effects of mitigation evolve due to population feedback. Figure is reproduced from 10.1101/2020.06.03.20120691.

Gamma-ray burst afterglows

Gamma-ray bursts (GRBs) are highly energetic explosions that were first observed by Project Vela, a military space project intended to detect terrestrial nuclear explosions. GRB light curves exhibit characteristic afterglow signatures that are thought to be related to the geometry and dynamical evolution of their underlying jets as they collide and interact with the surrounding material.

Schematic diagram of GRB fireball model that shows the evolution of the jet (launched from a central engine at T=0s) as shocks form and it interacts with the matter in its local neighborhood. Figure from Ghisellini 2001.

I produced numerical models of trans-relativistic blast waves using the relativistic hydrodynamic adaptive mesh refinement code JET. By comparing synthetic light curves from these models to observational data from the Swift-XRT mission, it is possible to infer the universal (meta-)distribution of parameters like the inclination angle between the GRB jet axis and the line of sight to observers on Earth. This parameter inference task was accomplished using the Markov chain Monte Carlo python package emcee.

Example posterior probability distribution for parameters (including plasma density, opening angle, and energy distribution parameters) describing GRB 110422A. The best-fit curve (black) is plotted over the Swift data (red points) in the panel at the top right. Figure from Ryan+ 2015.

5G wireless

I was an inaugural member of NYU WIRELESS, where I helped design and execute the first millimeter wave (at 28 GHz) channel sounding studies in New York City. After co-leading the experimental campaign, I produced models for the millimeter wave channel response of urban canyons (including characterizations of reflection and attenuation), developed statistical models of absolute power delay profiles for millimeter-wave channel simulators, and worked on multiple-input multiple-output link switching algorithms. I have been invited to present this research in a variety of sectors including academic conferences, board meetings, and government forums. This work led to my being a co-recipient of the 2015 Donald G. Fink award.

Left panel: small scale fading effects showing changes to the power delay profile as a function of position (in half-wavelength increments) for a 135 meter transmitter–receiver separation. Right panel: angular structure of received power (due to reflections off buildings) for 78 meter transmitter–receiver separation. Figures are reproduced from 10.1109/ACCESS.2013.2260813.
Map of selected transmitter and receiver locations in Manhattan near the New York University's Washington Square Park campus, showing locations of successful radio links. Figure is reproduced from 10.1109/ACCESS.2013.2260813.