MATERIALS DISCOVERY

Find the catalyst.
Cut the wait.

A spatial-adaptive active learning strategy identifies a Cu–RuO₂ catalyst that couples exceptional acidic OER activity with long-term durability.

177 mVOverpotential @ 10 mA cm⁻²
625 hStable operation in acid
78×Stability improvement
THE CHALLENGE

Optimize what is fast.
Then optimize what lasts.

Activity can be measured in minutes; durability testing can take hundreds of hours. Treating both objectives equally slows discovery.

This work links two targeted learning stages: it first finds low-overpotential recipes, then constructs a focused candidate space to search for stability.

19,200 → 785

A search space with focus

A conditional variational autoencoder concentrates the original recipe space into promising low-overpotential candidates.

23 / 11

Experiments / stability tests

The closed loop reaches the design target with only 11 long-duration durability measurements.

4,734 h

Testing time saved

At least 12 stability tests—and approximately 4,734 testing hours—are avoided during discovery.

RESULTS AT A GLANCE

A smarter route to durable performance.

The workflow narrows the question, the learning loop finds the answer, and characterization confirms the resulting Cu–RuO₂ catalyst.

Diagram of the spatial-adaptive active learning workflow
Two-stage spatial-adaptive learningBayesian optimization for activity is followed by CVAE-guided stability optimization.
Active learning and conditional variational autoencoder results
Learning the promising regionIterative experiments improve activity predictions before the CVAE refines the stability search space.
Catalyst stability and electrochemical performance results
Performance validatedCu–RuO₂ reaches 177 mV overpotential and remains stable for 625 hours in acidic conditions.
Microscopy, spectroscopy, and explainability results
Structure and insightMicroscopy, diffraction, spectroscopy, and model interpretation connect discovery to material evidence.