Skip to main content
Highlights

Fast Phase Prediction of Charged Polymer Blends by White-Box Machine Learning Surrogates

In this work, we developed a white-box surrogate model approach that expedites the the most computationally intensive element of the RPA simulation that relies only on two parameters that we predict with a parallel partial Gaussian process (PP-GP) surrogate model.

The white-box approach is over 100 times faster than direct simulation while still being over 99.9 % accurate in phase prediction based on only 50 training data points over the entire input space with 13 dimensions. Leveraging this tool, permits researchers to rapidly predict phase boundaries of charged polymers and efficiently screen the high-dimensional input space of charged polymers.

Ellis, Fang, Balzer, Quah, Shell, Fredrickson, Gu, Fast Phase Prediction Of Charged Polymer Blends By White-box Machine Learning Surrogates, Macromolecules 59 (2025) 202950–202959. DOI: 10.1021/acs.macromol.5c02482

Workflow of the fast white-box surrogate model, with physics-aware prediction and rapid screening of high-dimensional parameter space
Workflow of the fast white-box surrogate model, with physics-aware prediction and rapid screening of high-dimensional parameter space
Share:

See Original File

See the original highlight
Download
Publication

Fast Phase Prediction of Charged Polymer Blends by White-Box Machine Learning Surrogates

Macromolecules
View Publication

Materials Research Science and Engineering Center at UCSB

The NSF Materials Research Science and Engineering Center at UC Santa Barbara develops and sustains a productive, collaborative, and engaged community that drives a portfolio of transformative materials research and empowers a diverse workforce.