Simulator
Differometor
JAX-based autodifferentiable simulator for gravitational-wave detector design.
A physics experiment design competition for gravitational-wave detectors
Team
- Department for Computer Science, Faculty of Science, University of Tübingen, Tübingen, Germany
- Feyer, Tübingen, Germany
- Zuse School ELIZA, Darmstadt, Germany
- Institut für Festkörpertheorie und Optik, Friedrich-Schiller-Universität Jena, Jena, Germany
- Nikhef, National Institute for Subatomic Physics, Amsterdam, The Netherlands
- Institute for Quantum Information and Matter, California Institute of Technology, Pasadena, CA, USA
Submit optimization algorithms, not fixed designs. Each method tunes roughly 200 continuous parameters for a detector topology under a 4-hour evaluation budget using a differentiable simulator.
Central question
Can machine learning discover experimental designs that go beyond
human intuition while remaining physically meaningful
and experimentally constrained?
Design archive
30,000 high-quality designs
Released detector blueprints for learning, warm starts, and search.
Prize pool
EUR 25,000
Sponsored by SPRIN-D for the top-performing methods on hidden topologies.
What To Do
Three steps to compete
Step 01
Study the starter kit
Use the starter kit in the official GitHub repository, plus the simulator, dataset, and baselines, to understand the search space and evaluation setup.
Step 02
Design your algorithm
Tune roughly 200 continuous parameters for each topology within the 4-hour evaluation budget.
Step 03
Submit one zip folder
Submit a single Python class implementing the competition interface, one requirement file and any other additional files needed.
Submission
Key dates and format
Portal status
Submissions open
Public leaderboard updates begin in August. Final submission deadline: 15. October 2026.
What to submit
One algorithm class
Submit a single .py file, plus an optional requirements.txt.
How scoring works
Average best loss over 10 hidden topologies
Lower is better.