About Me

I am a Kavli Fellow at the Kavli Institute for the Physics and Mathematics of the Universe (Kavli IPMU), University of Tokyo. Previously, I was a postdoctoral researcher at Lawrence Berkeley National Laboratory and Princeton University.

My research connects astrophysical simulations, machine learning, and statistical inference to understand how structure emerges in the universe—from the intergalactic medium to galaxies and dark matter halos. I develop differentiable, learnable simulation frameworks that integrate data directly into the modeling process, allowing the universe itself to help refine our physical models.

I’m also deeply involved in educational outreach through Splash and Learning Unlimited, programs that empower university students to teach short, creative courses to local high school and middle school students. I’ve mentored and helped launch programs at UC Berkeley, UC Merced, Oxford University, Bard College, and Northwestern University.

Current Research Focus

1. Reconstructing the Large-Scale Structure of the Universe

The distribution of matter—visible and dark—encodes the history of structure formation. I build maximum-likelihood and differentiable reconstruction methods to infer cosmic density and velocity fields from the Lyman-α forest, spectroscopic galaxy surveys, and the CMB.

These efforts powerfully constrain the evolution of cosmic structure and bridge early-universe initial conditions with present-day galaxy environments.

2. Machine Learning and Differentiable Simulations

I develop GPU-accelerated, differentiable hydrodynamics and hybrid physical–ML systems that plug directly into inference pipelines. These learnable simulations adapt their physics to observations, enabling self-correcting, data-driven astrophysics.

These tools accelerate parameter inference and enable end-to-end differentiable astrophysical modeling, where cosmic structure formation can be jointly optimized with data.

3. Mapping Gas, Galaxies, and the Cosmic Web

Using tomography of the Lyman-α forest, I have reconstructed the three-dimensional cosmic web at redshifts z ≈ 2–3, revealing how intergalactic filaments feed forming galaxies.

This work underpins the cosmosTNG simulation suite (Byrohl + Horowitz et al. 2025), which evolves observed initial conditions into full hydrodynamical realizations of the COSMOS field, enabling one-to-one comparison between simulated and real galaxies.

4. Linking Observations and Fundamental Physics

The same frameworks are being extended to probe the Cosmic Microwave Background and Beyond-Standard-Model physics, such as primordial black holes and dark matter interactions.

Recent Projects and Collaborations

cosmosTNG — Constrained hydrodynamical simulations of the COSMOS field.
Prime Focus Spectrograph (PFS) — Galaxy Evolution Working Group, Large-Scale Structure Coordinator.
CLAMATO Survey — Principal developer of tomographic reconstruction framework and DR2 release.
Center for Data-Driven Discovery (CD³), University of Tokyo

Benjamin Horowitz

Who I work with...