| ---
|
| license: mit
|
| tags:
|
| - scientific-machine-learning
|
| - ai-for-science
|
| - physics-informed-machine-learning
|
| - agentic-ai
|
| - boiling
|
| - phase-change
|
| ---
|
|
|
| # Boiling Intelligence
|
|
|
| A minimal prototype of an autonomous AI physicist for phase-change mass-transfer discovery.
|
|
|
| ## v0.4
|
|
|
| The current synthetic benchmark demonstrates a closed scientific loop:
|
|
|
| 1. Maintain competing mass-transfer closures.
|
| 2. Design discriminating tests.
|
| 3. Query a hidden synthetic physical world.
|
| 4. Update evidence.
|
| 5. Detect model-class failure.
|
| 6. Infer a minimal constitutive correction.
|
| 7. Compile the correction into a new executable model.
|
| 8. Re-enter the revised model into the falsification loop.
|
|
|
| ### Example discovered closure
|
|
|
| \[
|
| M_3 = 0.8\,\Delta T + 0.00196515\,\Delta T^2
|
| \]
|
|
|
| The hidden synthetic law is used only to validate the autonomous discovery architecture.
|
|
|
| **This release does not claim discovery of a real boiling constitutive law.**
|
|
|
| ## Run
|
|
|
| ```bash
|
| pip install -r requirements.txt
|
| ollama serve
|
| ollama pull gpt-oss:20b
|
| python main.py
|
| ```
|
|
|
| ## Roadmap
|
|
|
| - Agentic scientific reasoning
|
| - Conservative multiscale recurrent neural operators
|
| - Basilisk phase-change simulations
|
| - Experimental ground-truth integration
|
| - Closed-loop physical model discovery |
