Journal
MRS COMMUNICATIONS
Volume 12, Issue 5, Pages 966-974Publisher
SPRINGER HEIDELBERG
DOI: 10.1557/s43579-022-00283-5
Keywords
Dielectric properties; Machine learning; Water; Simulation; Statistical methods
Categories
Funding
- Faculty Early Career Development Program of the National Science Foundation [DMR-1944211]
- U.S. DOE's National Nuclear Security Administration [DE-NA-0003525]
- Michigan Tech's doctoral finishing fellowship
Ask authors/readers for more resources
This study develops ensemble neural networks (ENN) as computationally fast surrogate models for Stockmayer fluid molecular dynamics (MD) simulations to determine the dielectric constants of polar solvents and NaCl solutions. The ENNs are trained using only 50-times less data compared to MD simulations, yet their predictions using batch normalization or bagging are in good agreement with the full MD results. The ENN methods are capable of extracting reliable values from statistically noisy data.
We develop ensemble neural networks (ENN) that serve as computationally fast surrogate models of Stockmayer fluid molecular dynamics (MD) simulations for determining the dielectric constants of polar solvents and NaCl solutions. The ENNs are trained using 50-times less data than is used to calculate the dielectric constants from MD simulations. The predictions of ENNs trained on this small amount of data and using batch normalization or bagging are in relatively good agreement with the full MD results. These ENN methods are thus able to extract reliable values from statistically noisy data.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available