MXene Catalyst Screening for Hydrogen Production

ISEF Category: Energy: Sustainable Materials and Design

Subcategory: Hydrogen Generation and Storage  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

Hydrogen is a clean fuel, but making it efficiently still takes smart catalysts. Think of a catalyst like a shortcut for a reaction, it helps the chemistry happen without being used up. In this project, you can use open materials data and Python to guess which MXene materials might work best. That gives you a real research question, not just a demo.

What Is It?

MXenes are a family of ultrathin materials made from transition metals and carbon or nitrogen. Picture stacked sheets you can peel apart into paper-thin layers. Their structure gives them lots of active surface area, so researchers study them as catalysts for the hydrogen evolution reaction, or HER. HER is the reaction that makes hydrogen gas from water.

Your job is to use existing data, not to synthesize the material. You will start with calculated properties from sources like the Materials Project and look for patterns that link material features to catalyst performance. In simple terms, you are asking, which material traits seem to go with easier hydrogen production? A regression model can help you turn many features into a predicted performance score.

Why This Is a Good Topic

This topic works well because you can test real hypotheses with public data and a clear outcome. You do not need a chemistry lab, but you still get to work with a problem tied to clean energy and catalyst design. You can learn data cleaning, feature selection, model evaluation, and how scientists compare materials before anyone makes them in the lab.

Research Questions

• How does MXene composition affect predicted hydrogen adsorption strength in a regression model?
• What is the effect of adding surface terminations such as O, F, or OH on predicted HER performance?
• Does including electronic features improve model accuracy more than using composition alone?
• To what extent can a simple Python regression model rank MXenes in the same order as published DFT values?
• Which feature set gives the lowest prediction error for MXene-based HER activity?
• How does training on one MXene subgroup compare with training on the full dataset for out-of-sample prediction?

Basic Materials

• Laptop or desktop computer with internet access.
• Python installed with Jupyter Notebook or Google Colab access.
• Public materials data from the Materials Project or related open datasets.
• Spreadsheet software for data cleaning and tracking variables.
• Free graphing tool or Python plotting library such as Matplotlib or Seaborn.
• Notes document for recording feature choices, model settings, and results.

Advanced Materials

• Access to the Materials Project API or downloadable dataset exports.
• Python environment with pandas, NumPy, scikit-learn, Matplotlib, and Seaborn.
• Optional descriptor tools for materials features such as matminer.
• A local machine with enough memory to handle larger tables and repeated model runs.
• Version control with Git for tracking code changes.
• ImageJ for checking any published figure data you may need to digitize from papers.

Software & Tools

• Python: Runs data cleaning, feature engineering, regression, and plotting for the screening workflow.
• Jupyter Notebook: Keeps code, notes, and figures together in one shareable file.
• scikit-learn: Builds and tests simple regression models for predicting catalyst performance.
• pandas: Organizes the materials table, handles missing values, and joins data sources.
• Matplotlib: Makes plots that compare predicted and actual HER metrics across MXenes.

Experiment Steps

1. Define the exact HER metric you will predict, and choose a dataset where that metric appears consistently.
2. Select a small set of material features that you can justify from chemistry, such as composition, surface termination, or electronic descriptors.
3. Clean the dataset so every row represents one comparable MXene entry and every missing value has a clear plan.
4. Split the data into training and test groups so you can check whether your model predicts new examples.
5. Compare a few simple regression models, then decide which one balances accuracy and interpretability.
6. Test whether your top features and ranked candidates match known trends from published DFT studies.

Common Pitfalls

• Mixing values from different papers or databases, which can blur the meaning of the HER target.
• Using too many features for a small dataset, which makes the model memorize noise instead of learning trend.
• Forgetting to check whether all rows use the same definition of adsorption energy or activity metric.
• Ranking MXenes by raw prediction alone, which hides uncertainty and makes weak predictions look certain.
• Ignoring class imbalance or duplicate entries, which can inflate accuracy and make validation look better than it is.

What Makes This Competitive

A stronger project does more than fit a line to a dataset. You can compare several feature sets, test how stable the rankings are, and check whether your model still works on unseen MXene families. Strong entries explain why the model works, not just which score won. If you tie the predictions back to known chemistry and discuss uncertainty clearly, your project starts to look like real materials screening.

Project Variations

• Screen MXene catalysts for oxygen evolution instead of hydrogen evolution, and compare whether the same features still matter.
• Use classification instead of regression, and ask whether a model can separate high-performing MXenes from low-performing ones.
• Add published experimental data to the DFT table, then test whether theory-only features still predict real catalyst performance.

Learn More

• Materials Project: Search the public database for calculated materials properties, then look for MXene entries and related DFT descriptors.
• PubMed: Search review articles on MXenes and hydrogen evolution reaction catalysis to understand current research questions.
• NIST Materials Data Repository: Explore open materials datasets and learn how scientists organize computational results.
• MIT OpenCourseWare: Use introductory machine learning and materials science courses to build background on regression and descriptors.
• scikit-learn User Guide: Read the free documentation for regression, model evaluation, and train-test splits.