Radical Artificial Intelligence for Multiphase Environmental Systems
Abstract
Accurate prediction of the future given a set of chemical reactants is a desirable capability with broad application. Robust estimation of chemical speciation with models at long reaction times is currently too difficult for present computer science techniques due to error propagation over multiple steps and high computational costs. This is true for single-phase chemical systems, and even more challenging for multiphase and interfacial systems. For example, evolution of organic compounds in multiphase systems involves simultaneous sequential and parallel chemical corridors in addition to phase transitions. Artificial intelligence and machine learning lend themselves to chemical discovery in this area, in part, because traditional approaches are difficult when mechanistic pathways are highly branched and convoluted. While the field has some understanding of the multiphase organic phenomena in relatively simple systems, little is known under realistic, non-ideal conditions. Statement of Work: We propose an interdisciplinary chemistry/computer science project to elucidate and explain measured data for multigenerational oxidation systems at the forefront of environmental multiphase chemistry research to provide insights to the critical factors that control and improve predictive skill and seek to answer: To what extent can we achieve human expert skill to predict multiphase chemical reactions for environmentally real systems with Artificial Intelligence? Objectives: 1.) instruct a deep learning system with radical training reactions, 2.) combine radical and polar rule-based chemistry for single step reactions in a deep learning system, 3.) build on the single step approach for combined radical and polar chemistry in multi-generation oxidation, 4.) test and evaluate the deep learning system at all stages of model development. Methods: We propose to use Reaction Predictor, a rule- and machine learning-based system for predicting chemical reactions using deep learning. Reactions are predicted at the level of elementary mechanistic steps that can be chained together to yield complex global reactions. Significance: Multiphase chemistry defines all of the physicochemical transformations between states of matter on scales ranging from nanoseconds to millennia, subatomic particles to solar systems. We cannot predict beyond a few steps in simple systems, and rarely for a non-ideal system. From a chemical perspective, everything is a multiphase process. Evolution of primordial gas and dust, and mass exchange between the phases of the interstellar medium defines the Milky Way as we understand it. Chemical reactions and phase transitions within and between the atmosphere, biosphere, hydrosphere, and pedosphere/lithosphere defines all cycling in the Earth system. LifeÕs very metabolism is dependent on multiphase chemistry and the mass transfer between phases.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 09, 2020
- Source ID
- W911NF2010172
Entities
People
- Ann Marie Carlton
Organizations
- Army Contracting Command
- United States Army
- University of California, Irvine