Research

Splendors of elemental strife; Smit suns that startle back the gloom; New light whose tale of stellar doom Fares to uncomprehending life; – George Sterling, The Testimony of the Suns, 1907

The transient sky is diverse and ever-changing. Supernovae, having been observed and recorded by humanity for the whole of our species’ history, are among the most prolific phenomena observed in the night sky. These are magnificent explosions following stellar death. They are, literally, the last testimony of a dying star. It is our charge as practitioners of stellar astrophysics to interpret these testimonies: what is the star trying to tell us? What can we learn about its life, death, and environment? My work involves deploying and developing computational models to interpret these testimonies.

Open Source Simulation Codes

Scientific software poses a host of unique scientific, engineering, and collaborative questions. How do we create software that is robust, accurate, and capable of pushing the frontiers of science? How do we create scientific software that can survive into the future, answering not only the questions of today but those of tomorrow? Moreover, as we move into the era of exascale computing, we need modern simulation codes that can take full advantage of these architectures. This poses a significant engineering challenge as the number of programming models continues to grow. Ultimately, we seek to empower users to drive scientific innovation. The answer lies in research software engineering: the union of domain knowledge, numerical methods, and modern software development practices. I am a developer for a number of open source scientific codes working to answer these questions and more.

Athelas is being developed to simulate the electromagnetic emission from transients, primarily supernovae. Athelas evolves Lagrangian hydrodynamics coupled to a two moment radiation transport model. It leverages a discontinuous Galerkin spatial discretization of the fluid and radiation fields to acheive robust solutions Time integration is implemented with high-order, fully coupled implicit-explicit (IMEX) methods – no operator splitting is required. at high order accuracy. Athelas includes Saha ionization coupled to the equation of state, a radioactive decay network for nickel heating, and can artificially drive explosions. In the near future it will incorporate a Rayleigh Taylor mixing model. Athelas includes a very flexible package system for building up the partial differential equations being solved, automatically threading physics into the IMEX integrator, making it very simple to add new physics. Athelas includes the capability to map outputs from the MESA stellar evolutionary code into its input format to make robust predictions.

Phoebus is developed to tackle a range of problems in relativistic astrophysics, with emphasis on core-collapse supernovae, neutron star mergers, and black hole accretion. It is developed from the ground using a GPU-first development strategy. To facilitate performance portability, Phoebus is built on Kokkos, a parallelization abstraction layer that allows Phoebus to run on any GPU or CPU architecture by simply changing compile-time options.

Thornado is a GPU-capable code utilizing a discontinuous Galerkin phase-space discretization developed primarily for modeling core-collapse supernovae. By using discontinuous Galerkin methods for both hydrodynamics and neutrino radiation transport, thornado can achieve high order accurate solutions for both the fluid and radiation fields, capturing the complex flows that assist in shock revival, and ensure realizability of the fermionic neutrino distribution functions. Thornado provides neutrino transport capabilities to the multiphysics code Flash-X.

I contribute to a number of other frameworks that support downstream science production codes such as the parthenon adaptive mesh refinement framework and the singularity-eos performance portable equation of state library.

On my GitHub you may find software that I have produced, including codes for protytyping and exploring methods.

Discontinuous Galerkin Finite Element Method

Discontinuous Galerkin methods are a very promising method for simulating astrophysical phenomena. They can be thought of as the high-order love child of finite volume and finite element methods. Similar to finite elements, the solution is represented as a basis expansion. However, unlike finite elements, these basis representations may be discontinuous across cells. Neighboring cells are connected via numerical fluxes – Riemann solvers – identical to finite volume methods. By evolving a high-order approximation of the solution on each cell instead of cell averages we avoid the need for complicated reconstruction steps. As a result, the computational stencil is independent of the order of accuracy — only nearest-neighbor communication is needed, making them especially parallelizable. Discontinuous Galerkin methods have a number of desirable qualitities, including improved angular momentum conservation, more natural capturing of asymptotic limits in radiation transport, and polyunomial order adaptivity in space. They are high resolution shock capturing schemes that are conservative of the cell average in the finite volume sense.

Using Realistic Explosion Models To Interpret Core-Collapse Supernovae Observations

Observations of core-collapse supernovae (CCSNe) reveal a wealth of information about the dynamics of the supernova ejecta and composition but tell little of the progenitor star without invoking a theoretical model. Until recently, one-dimensional (1D) theoretical CCSN models did not include a robust treatment of the core physics, resorting instead to artificial thermal bomb explosions. These simplified methods input arbitrary explosion energy into a progenitor to induce an explosion.

I use a new model for driving turbulence-aided neutrino-driven core-collapse supernovae in 1D, which contains a high-fidelity treatment of the neutrino physics while also accounting for turbulence and convection, which can reproduce properties of 3D simulations. Moreover, our light curve features agree very well with a population of observed Type IIP supernovae. With this, I can begin to connect explosion and progenitor properties for a realistic population of CCSNe to apply to observed CCSNe. This is especially exciting as LSST and other next-generation surveys prepare to collect unprecedented amounts of data.

The use of realistic explosion models is important for understanding observables and connecting back to the progenitor star. It is becoming clear that explosion properties inferred from light curves are not unique (e.g., Goldberg 2019) — many progenitor and explosion energy combinations can reproduce a light curve. Using a realistic explosion model — where the explosion energy is determined naturally by properties of the progenitor and neutrino physics — could help to reduce this problem by excluding progenitor-explosion energy combinations that cannot be achieved by more physical models.

An initial paper has been published where we explore this method for a suite of over 100 progenitors. We find excellent bulk agreement with observations of other Type IIP supernovae, notable differences in estimated progenitor properties when trying to infer properties from observed light curves, and a strong, linear relationship between plateau luminosity and iron core mass.

Following up on this, we used these results with real observations to infer iron core masses for a population of SNe. In this paper, we apply Bayesian inference methods to connect the inferred iron core mass distribution to ZAMS mass properties, finding evidence of high mass red supergiant progenitors.

Equation of State Dependence of Core-Collapse Supernova Observables

The outcomes of core-collapse supernova (CCSN) simulations can depend sensitively on the nuclear physics of dense matter through the equation of state (EOS). In the simulation of core-collapse supernovae, the EOS of nuclear matter is included through phenomenological models and encoded in an EOS table for use in our simulations as a relationship between the thermodynamical variables. Due to limitations in our understanding and in computational feasibility, there are many different models for the nuclear force, resulting in many different tables that we may include in our simulations. Therefore, a quantitative understanding of how different EOS tables affect the outcome of core collapse is crucial to our ability to make predictions. 1D CCSN simulations are ideal tools for understanding this sensitivity, as we can run thousands of 1D simulations to explore the parameter space meaningfully. Using a new model for creating physical explosions in 1D, I explored the sensitivity of CCSNe to variations in input nuclear physics by performing a population study using nine open-source EOS tables and 138 progenitor stars with ZAMS masses ranging from 9 to 120 solar masses. I tested the sensitivities of the observable signals to the nuclear equation of state and explored correlations of the signals with fundamental nuclear physics quantities, such as the symmetry energy and effective nucleon mass.

Prospects for High Energy Follow-up Studies of Gravitational Wave Transients

Many of the most violent and energetic events in the cosmos, in particular the merger of compact objects and core-collapse supernovae, are sources of gravitational waves and are also believed to be connected with Gamma Ray Bursts. Joint observations of electromagnetic and gravitational wave signals will provide an ideal opportunity to study the physics of these transient events and their progenitors. In particular, gamma-ray observatories such as Fermi, coupled with precise sky localization, will be crucial to observe the high-energy electromagnetic counterparts to gravitational wave signals. We constructed joint binary neutron star and gamma-ray burst detection rate estimates using an analysis pipeline and report on the results of this analysis. Moreover, I extended the analysis to include a catalog of real galaxies appropriate for such a follow-up study (the GLADE Catalog).

Other Research

• The Nature of Supernova Shock Revival: After the onset of core collapse, the shock responsible for tearing the star apart and driving the explosion runs out of energy and stalls. The exact nature of how the shock is revived is still a matter of active research. I developed data analysis tools to investigate the relative contributions of neutrino-driven convection and the standing accretion shock instability to shock revival.