Ab initio molecular dynamics (AIMD) simulation of chemical reactions requires accurate force calculations to propagate the reaction system in time (i.e. a trajectory). An overwhelming majority of the computation is dedicated to force calculations with quantum chemistry methods, which takes place at every frame of the trajectory, and the result of force calculations is thrown away immediately after. A variety of methods, such as Hessian update and machine learning, have been developed and implemented to reduce the number of force calculations. In this presentation, we will show a novel active learning protocol that predicts the forces from interpolating previously sampled trajectories. Unlike other methods, our active learning protocol does not require Hessians to calculate interatomic forces nor seeks a global functional form of forces with respect to configurations of the system. Instead, it constructs a database as simulations go, attempts an interpolation of forces with the assessment of its reliability, and is able to fall back onto the quantum chemistry calculation whenever the interpolation is deemed risky. Our method has been proven mathematically sound, and a framework has been developed for immediate implementation in existing AIMD programs. The proposed method &lt;!-- The body of an abstract is usually between 200 to 250 words. Yours seems too short --&gt;has been applied to reaction simulations of HBr+ + CO2 HOCO+ + Br and has demonstrated its validity by reducing the number of quantum force calculations by half, without sacrificing the outcome of the trajectories.

Active learning algorithm to reduce force calculations in ab initio molecular dynamics

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.