AMS Ideas in Action: Modeling Surface Tension in Binary Liquid Mixtures

In this AMS Ideas in Action post, we share a standalone Python notebook for modeling the surface tension of binary liquid mixtures using experimental data from the NIST data package on viscosity and surface tension of binary mixtures.

Surface tension is an important property in formulation science, coatings, wetting, emulsions, separations, and process development. For binary liquid mixtures, it can depend strongly on composition, temperature, and the molecular interactions between the two components. This makes it an interesting target for fast, data-driven property modeling.

With AMS Ideas in Action, we want to share applied notebooks that can be explored, modified, extended, and discussed with SCM experts. This example shows how experimental mixture data, molecular descriptors, and machine-learning models can be combined into a practical workflow for rapid prototyping.

From NIST Data to an Executable Notebook

The notebook uses a curated table derived from the NIST binary-mixture surface-tension data, while also keeping the raw NIST files available for provenance and inspection.

The workflow focuses on neutral binary systems and builds several descriptor blocks, including:

• composition and temperature features;
• RDKit molecular descriptors;
• COSMO-based SG1/FastSigma sigma-profile moments;
• PropPred pure-compound properties such as density, viscosity, vapor pressure, and boiling point.

These descriptors are then compared in a light machine-learning benchmark using a tree-based ensemble model. The split is grouped by binary pair, so the test set contains liquid pairs that were not seen during training.

Why This Is Interesting for Mixture-property Screening

For industrial mixture design, experimental data is often valuable but unevenly distributed across chemical space. A notebook like this can help explore questions such as:

• Which descriptors are useful for binary-mixture surface tension?
• Can pure-compound properties and molecular descriptors support mixture-property predictions?
• How well does a model generalize to binary pairs outside the training set?
• What would be needed to adapt the workflow to a proprietary formulation dataset?
• Where could COSMO-RS, SG1 descriptors, or other AMS tools add value in a more detailed follow-up study?

This is the kind of rapid prototyping workflow where notebooks are especially useful: fast to evaluate, transparent in each step, and highly customizable for different datasets, descriptors, target properties, or modeling choices.
From Literature Idea to Executable Notebook

Discuss Your Mixture-property Modeling Challenge with us

Are you working on liquid mixtures, formulations, solvent systems, coatings, wetting, or thermophysical property prediction?

We invite you to schedule a discussion with SCM experts. Together we can look at your target property, available data, chemistry space, and the level of simulation or validation that would be appropriate for your research question.

Contact us to discuss how AMS simulations, COSMO-based descriptors, and machine-learning workflows could support your mixture-property modeling project.

New to Python notebooks? See our short introduction to running AMS notebooks.