Peter Nugent

Agent-based modeling & epidemiology

ExaEpi is an agent-based epidemiological code built on the AMReX framework, able to simulate the United States agent-by-agent on GPU-accelerated exascale machines. It is the simulation engine of EMERGE, a DOE/ASCR multi-lab project on robust, generalizable epidemiology — including large ensemble campaigns on Aurora and Perlmutter, coupling to CMIP6 climate and weather projections, surrogate and reduced-order modeling, and venue-by-venue validation against LANL's EpiCast.

Critical minerals & supply-chain modeling

pyExaMINE (Python ExaScale Minerals & Infrastructure Network Evaluation) is an agent-based model of worldwide critical-mineral supply chains at facility resolution — individual mines, processors and recyclers for lithium, nickel, cobalt, manganese and platinum, with shipments routed through the real maritime chokepoints (Hormuz, Suez, Malacca, Panama, the Cape) and price dynamics driven by embargoes, chokepoint crises and material substitution.

AI for science — the Genesis Mission

Two projects selected in the DOE Genesis Mission RFA:

AHPI, the Agentic HPC Pipeline Initiative, will build AI workflows that enable U.S. manufacturers to run powerful DOE research codes on commercial cloud platforms.

HERALD, High-throughput ECI Review and Agentic Legacy Document processing, will build an AI platform to vet decades of secure government research, unlocking it for the commercial industry to accelerate nuclear innovation.

Time-domain surveys: PTF, ZTF and LS4

I was PI of the Type Ia supernova project within the Palomar Transient Factory (2008–2014), which built the real-time subtraction and machine-learning pipeline at NERSC that found SN 2011fe within hours of explosion. That work carried into the Zwicky Transient Facility — transient discovery at scale, plus searches for microlensing black holes and for the tidal-disruption and AGN variability signatures of massive black-hole mergers — and now into LS4, the La Silla Schmidt Southern Survey, where I am Project Scientist and Deputy Director.

Supernova physics

The discovery and observation of supernovae of all types, with the goal of understanding the physics of their explosions, their progenitor systems, and their nucleosynthesis products. Spectrum synthesis is the main tool: 1-D non-LTE with PHOENIX, and 3-D LTE Monte Carlo with SEDONA. Recent threads include sub-Chandrasekhar and super-Chandrasekhar Type Ia models, circumstellar interaction, and the interior structure of dying stars.

Supernova cosmology

Measuring the cosmological parameters using supernovae as distance probes — not only Type Ia but also Type II-P, via the spectral-fitting expanding atmosphere method, and strongly lensed supernovae, whose time delays measure H₀ independently of the distance ladder. The K-correction and extinction-correction machinery underpinning much of this is what the spectral templates on this site are for.

Computational & data-intensive science

I am always looking for better and faster ways to solve an interesting problem: GPU porting and performance at scale, hybrid-parallel deep learning on large 3-D data, surrogate models that stand in for expensive simulations, real-time pipelines that turn telescope pixels into alerts, and the workflow plumbing that makes ensembles of 100,000 runs tractable. I have also worked on N-body + hydrodynamic simulations of the Lyman-α forest with the Nyx code.