NequIP Proton Hopping in Carbon Nanotubes

ISEF Category: Chemistry

Subcategory: Computational Chemistry  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A single proton can move through water faster than you might expect, almost like a baton passing through a crowd. Inside a carbon nanotube, that motion can change again because the water is squeezed into a tiny channel. Your model can test how tube diameter changes that hop-and-pass process. That makes this project part chemistry, part physics, and part machine learning.

What Is It?

In bulk water, a proton does not just drift like a tiny ball. It often moves by the Grotthuss mechanism, which means the proton jumps from one water molecule to the next as bonds rearrange. Picture a line of people passing a message down the chain. The message moves faster than any one person walking.

Inside a carbon nanotube, water becomes confined. The tube diameter can change how water molecules line up, how many hydrogen bonds they form, and how easily the proton can hop. An equivariant neural-network potential, such as NequIP or MACE, is a machine-learning model that learns atomic interactions while respecting rotation and translation symmetry. That gives you a way to simulate many atoms more quickly than a full quantum calculation, while still keeping important chemistry in the model.

Why This Is a Good Topic

This is a strong science fair topic because you can vary one clear factor, nanotube diameter, and measure how it changes proton transport. The project connects to fuel cells, membrane design, and water in nanoscale channels. You can learn molecular simulation, model training, validation, and data analysis without needing a wet lab. The question is narrow enough for a fair, but rich enough to support real research.

Research Questions

• How does CNT diameter affect the rate of proton hopping in confined water?
• What is the effect of CNT diameter on hydrogen-bond network order around the proton?
• Does an equivariant neural-network potential predict proton transfer pathways more accurately than a standard force-field model?
• To what extent does water orientation inside a CNT change as diameter decreases?
• Which CNT diameter gives the highest effective proton diffusion coefficient under the same simulation conditions?
• How does proton localization time change with CNT diameter?
• What is the effect of adding a defect or charge to the CNT wall on proton hopping?

Basic Materials

• A laptop or desktop computer with a modern multi-core processor.
• Access to a Linux environment or WSL for molecular simulation tools.
• Python 3 with NumPy, Matplotlib, pandas, and Jupyter Notebook.
• An open-source molecular dynamics package such as LAMMPS or ASE-compatible tools.
• An open-source machine-learning potential framework such as NequIP or MACE.
• Training and validation datasets from published water or proton-transport simulations.
• A digital notebook for tracking model settings, dataset splits, and results.
• Cloud storage or an external drive for simulation outputs and checkpoints.

Advanced Materials

• Access to a university Linux cluster or GPU workstation.
• A curated ab initio molecular dynamics dataset for water, protonated water, or carbon nanotube interfaces.
• NequIP or MACE source code with GPU support.
• PyTorch with CUDA support for model training.
• ASE for building nanotube and water configurations.
• LAMMPS, i-PI, or another molecular dynamics engine for production runs.
• MDAnalysis or MDTraj for trajectory analysis.
• VMD or OVITO for visualizing trajectories and hydrogen-bond structure.
• A high-quality reference method, such as DFT data, for validation and comparison.

Software & Tools

• Python: Runs analysis scripts, statistical tests, and custom plots for transport metrics.
• Jupyter Notebook: Helps you document model choices, figures, and results in one place.
• ASE: Builds nanotube geometries and handles atomic structures for simulation inputs.
• NequIP: Trains an equivariant neural-network potential that respects molecular symmetry.
• MACE: Trains a similar equivariant potential and provides a strong comparison model.
• MDAnalysis: Measures proton trajectories, hydrogen-bond lifetimes, and diffusion-related outputs.

Experiment Steps

1. Define the exact transport metric you will measure, such as hopping rate, residence time, or diffusion coefficient.
2. Select a small set of CNT diameters that cover a meaningful change in confinement without changing the chemistry of the wall.
3. Choose a reference dataset and decide how you will split training, validation, and test structures to avoid leakage.
4. Build a validation plan that compares your model against a higher-level reference for proton transfer and water structure.
5. Plan the analysis pipeline that turns raw trajectories into comparable numbers across tube sizes.
6. Predefine controls for temperature, water count, and proton state so diameter remains the main variable.

Common Pitfalls

• Training on structures that are too similar to the test set, which makes the model look accurate without real predictive power.
• Comparing CNTs with different water densities, which confounds diameter effects with crowding effects.
• Using only one random seed, which hides how unstable the neural-network potential may be.
• Measuring proton motion from noisy raw coordinates without defining a clear hopping criterion.
• Skipping model validation against a higher-level reference, which can leave you with a fast model that misses the chemistry.

What Makes This Competitive

A class-level project might show a trend. A stronger project explains why the trend happens. You can raise the level by testing more than one model, reporting uncertainty across random seeds, and using a careful validation set that the model never saw during training. Strong entries also compare transport with structure, so you connect hopping speed to water orientation, hydrogen bonding, and confinement.

Project Variations

• Compare proton hopping in armchair versus zigzag CNTs at the same diameter to see whether wall chirality matters.
• Test how proton transport changes when you replace pure water with slightly acidic water or hydronium-rich initial states.
• Analyze whether a simpler force-field model and an equivariant neural-network potential disagree most at small CNT diameters.

Learn More

• MIT OpenCourseWare, Molecular Dynamics and Statistical Mechanics: Search MIT OpenCourseWare for courses on molecular dynamics, statistical mechanics, and atomistic simulation.
• NIH PubChem: Look up molecular structures, including water, hydronium-related species, and carbon nanomaterials, for background and naming conventions.
• NIST Chemistry WebBook: Use it for thermodynamic and molecular property references when you compare simulation outputs to known values.
• PubMed: Search review articles on the Grotthuss mechanism, proton transport in water, and water in carbon nanotubes.
• Google Scholar: Search for recent papers on NequIP, MACE, and proton transport in confined water, then filter for review or method papers.
• Journal of Chemical Physics: Read peer-reviewed articles on atomistic simulation methods, proton transfer, and machine-learning potentials.