English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Machine learning force fields (MLFFs) are gradually evolving towards enablingmolecular dynamics simulations of molecules and materials with ab initioaccuracy but at a small fraction of the computational cost. However, severalchallenges remain to be addressed to enable predictive MLFF simulations ofrealistic molecules, including: (1) developing efficient descriptors fornon-local interatomic interactions, which are essential to capture long-rangemolecular fluctuations, and (2) reducing the dimensionality of the descriptorsin kernel methods (or a number of parameters in neural networks) to enhance theapplicability and interpretability of MLFFs. Here we propose an automatizedapproach to substantially reduce the number of interatomic descriptor featureswhile preserving the accuracy and increasing the efficiency of MLFFs. Tosimultaneously address the two stated challenges, we illustrate our approach onthe example of the global GDML MLFF; however, our methodology can be equallyapplied to other models. We found that non-local features (atoms separated byas far as 15$~\r{A}$ in studied systems) are crucial to retain the overallaccuracy of the MLFF for peptides, DNA base pairs, fatty acids, andsupramolecular complexes. Interestingly, the number of required non-localfeatures in the reduced descriptors becomes comparable to the number of localinteratomic features (those below 5$~\r{A}$). These results pave the way toconstructing global molecular MLFFs whose cost increases linearly, instead ofquadratically, with system size.

Towards Linearly Scaling and Chemically Accurate Global Machine Learning Force Fields for Large Molecules