Nathaniel Starkman

Nathaniel Starkman

Brinson Prize Fellow · Postdoc @ MIT Kavli Institute for Astrophysics

I am a computational astrophysicist researching dark matter — what it is, and how it shapes galaxies — mostly by using stellar streams to constrain their gravitational potentials, in the Milky Way and far beyond it with large-scale surveys.

When a star cluster is torn apart by its host galaxy, the debris can trace a long, thin stream. Properties of the stream — its path, shape, width, etc — are all sensitive to properties of the host, including its dark matter. I develop and apply novel methods to study these streams and learn about the dark matter; I also build much of the software powering these methods and analyses. I am also a core developer of Astropy, helping to power all of astronomy.

Background

2024 – 2027Postdoctoral Associate, MIT Kavli Institute for Astrophysics and Space ResearchBrinson Prize Fellowship (2024–2027).
2021 –Coordinator & Core Developer, AstropyRole: Member, Core Developer Team; lead maintainer for Cosmology, co-maintainer of Units.2025–2028: Coordination Committee (executive committee).2025–2026: Strategic Planning and Organizing Committee.Funding: Astropy Cycle III — Cosmology and Quantity 2.0.Team awards: 2025 AAS Lancelot M. Berkeley–New York Community Trust Prize, the 2023 IOP Publishing Top Cited Paper Award and the 2022 ADASS Prize for an Outstanding Contribution to Astronomical Software.
2018 – 2024PhD in Astronomy & Astrophysics, University of TorontoNSERC CGS-D Fellow 2020–2023. NSERC CGS-M Fellow 2019–2020.Thesis: Charting Stellar Streams of the Milky Way
2014 – 2018B.S. in Mathematical Physics and Astronomy, Case Western Reserve UniversitySumma Cum Laude.
Core developer · Coordination Committee · Strategic Planning

The community core package for astronomy in Python. The collaboration received the 2025 AAS Lancelot M. Berkeley Prize, a 2023 IOP Publishing Top Cited Paper Award and the 2022 ADASS software prize.

Orbit integration, potentials and stream generation — GPU-accelerated and fully differentiable.

Multiple dispatch for Python, with type-hint-driven method resolution — the dispatch layer under quax and unxt.

Vectors, frames and transformations — differentiable, and unit-aware via unxt.

Unit-aware quantities that survive jit, grad and vmap.

Custom array-ish types that work with JAX primitives — the substrate the rest of the stack builds on.