Combining High-Throughput Experiments and Active Learning to Characterize Deep Eutectic Solvents.

Area of Science:

• Physical Chemistry
• Materials Science
• Computational Chemistry

Background:

• Deep eutectic solvents (DESs) offer high tunability through precursor and composition variations.
• Characterizing DES physicochemical properties across wide ranges is experimentally intensive and hinders application development.
• Viscosity reduction via solvent addition is crucial for large-scale DES applications.

Purpose of the Study:

• To develop active learning (AL) methodologies using Gaussian processes (GPs) to minimize experimental effort in DES characterization.
• To explore the reduction of DES viscosity as a case study for AL application.
• To accurately predict DES viscosity using GPs and reduce data requirements.

Main Methods:

• High-throughput experimental screening of nine ternary DESs.
• Training Gaussian processes (GPs) to predict DES viscosity based on composition and temperature.
• Utilizing GP uncertainty estimates within an AL framework to guide data acquisition.

Main Results:

• GPs accurately predicted the viscosity of complex DES mixtures.
• The AL framework significantly reduced the number of required experimental data points.
• Accurate viscosity models were achieved with as few as five independent data points for many systems.

Conclusions:

• AL methodologies based on GPs effectively minimize experimental effort for DES characterization.
• This approach accelerates the design and scalability of DES-based applications.
• The study demonstrates a significant reduction in data requirements for accurate physicochemical property prediction.