Particula: Predictive Aerosol Research Through Modular Design and AI Integration

KYLE GORKOWSKI, Naser Mahfouz, Wayne Chaung, Los Alamos National Laboratory

     Abstract Number: 351
     Working Group: Advancing Aerosol Science through Data Analysis Tools

Abstract
Aerosol science is entering a new stage where individual investigators can pair laboratory data with numerical models as quickly as they collect it. Facilitating this moment is Particula (par-TICK-yoo-luh). This open‑source Python package unifies experimental and modeling workflows and plugs directly into large‑language‑model aides. It contains interchangeable particle‑resolved, moving‑bin, and bulk‑speciated solvers for Kelvin activation, condensation, Brownian and gravitational coagulation, turbulence effects, wall loss, and gas–particle partitioning.

A single script turns cloud-chamber or field data into a digital twin—matching activation spectra, and tracking composition‑resolved coagulation to reveal how gravitational collision–coalescence drives rapid mixing‑state homogenization. The same workflow handles chamber diagnostics and sensitivity analyses.

Particula’s “WARMED” ethos—Write, Agree, Read, Modify, Execute, Debug—keeps the API lean but transparent: prototype with pure NumPy/SciPy, then swap in compiled back‑ends without rewriting user code. Two OpenAI helpers cut the learning curve: Particula Assistant (GPT‑4o‑mini) answers code‑navigation queries, while Particula Reasoning drafts new simulations from plain language input.

Next on the roadmap: activity‑coefficient thermodynamics, GPU kernels, and an autonomous agent that returns analysis and figures. Everything is on GitHub (github.com/uncscode/particula), and every contributor earns co‑authorship on Particula manuscripts. By merging modular numerics with generative‑AI guidance, Particula lowers the barrier to predictive aerosol research and lets small, agile teams solve problems.