We present a novel algorithm for learning reaction networks from concentration time series in an unsupervised manner. Simulations of atomistic models of hydrogen oxidation do not require prior knowledge of possible mechanisms or intermediates. Here in, we use data from molecular dynamics simulations hydrogen oxidation to demonstrate analysis with our algorithm. We also analyze data from a trajectory within the basin of attraction of the Oregonator limit cycle. We demonstrate that stoichiometry and kinetics (reaction rates) can be obtained from Koopman modes, eigenvalues, and eigenfunctions. Reactions that occur consistently, and the high dynamical relevance (persistent reactions), are learned in an unsupervised manner; our persistent reaction computation compresses information contained in computed stoichiometric vectors while rejecting components due to numerical noise. The first persistent reaction computed for hydrogen oxidation, for example, is the corresponding overall reaction. Using this reaction and its corresponding Koopman eigenvalues, we define a data-driven ignition time which correlates well with an empirical ignition time. Overall, this unsupervised method is a potential means to learn bulk chemistry for a large variaty of reactions using only concentration time series as input.

Unsupervised construction of reaction networks from concentration time series via Koopman mode decomposition: Oscillatory and non-oscillatory reactions

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.