newclid 3.0.1

Solver for plane geometric problems.

Newclid is an open-source, easy-to-use fast solver for plane geometry problems.

Installation

Using uv (recommended)

Follow uv installation instructions

uv add newclid[yuclid]

Using pip

pip install newclid[yuclid]

Building from source

Follow uv installation instructions

git clone https://github.com/Newclid/Newclid.git
cd Newclid
uv sync

If you run into issues to build yuclid, you might need to set an environment variable like CXX=/usr/bin/g++-14, try adding environment variable that are present in .env. If you still have issues, submit an issue.

Quickstart

To simply solve a problem using Newclid, use the command line.

For example with a JGEX problem:

newclid jgex --problem-id orthocenter_consequence_aux --file ./problems_datasets/examples.txt

Or with a ggb problem:

newclid ggb --file ./notebooks/ggb_exports/incenter.ggb --goals "eqangle C B B D B D B A"

See other command line interface options with:

uv run newclid --help
uv run newclid jgex --help
uv run newclid ggb --help

For more complex applications, use the Python interface. Below is a minimal example to build a problem setup from a JGEX string, then solve it:

from newclid import GeometricSolverBuilder, GeometricSolver
import numpy as np

# Set the random generator
rng = np.random.default_rng()

# Build the problem setup from JGEX string
problem_setup = JGEXProblemBuilder(rng=rng).with_problem_from_txt(
  "a b c = triangle a b c; "
  "d = on_tline d b a c, on_tline d c a b; "
  "e = on_line e a c, on_line e b d "
  "? perp a d b c"
).build()

# We now build the solver on the problem
solver: GeometricSolver = GeometricSolverBuilder().build(problem_setup)

# And run the solver
success = solver.run()

if success:
    print("Successfuly solved the problem! Proof:")
    solver.write_proof_steps()
else:
    print("Failed to solve the problem...")

print(f"Run infos {solver.run_infos}")

In the notebooks folder you will find more tutorials. You can also check tests to see some more advanced examples of scripts using the Python interface.

Documentation

See the online documentation for more detailed information about Newclid.

Contributing

1. Clone the repository

git clone https://github.com/Newclid/Newclid.git
cd Newclid

2. Install uv

Follow installation instructions

3. Install as an editable package with dev requirements

uv sync

4. Install pre-commit and pre-push checks

pre-commit install -t pre-commit -t pre-push

5. Run tests

pytest tests

About Newclid

Newclid is a successor to AlphaGeometry, introduced in this early 2024 Nature paper: Solving Olympiad Geometry without Human Demonstrations. whose original codebase can be found here.

If you found Newclid useful, please cite us as:

@article{newclid2024sicca,
  author  = {Sicca, Vladmir and Xia, Tianxiang and F\'ed\'erico, Math\"is and Gorinski, Philip John and Frieder, Simon and Jui, Shangling},
  journal = {arXiv preprint},
  title   = {Newclid: A User-Friendly Replacement for AlphaGeometry with Agentic Support},
  year    = {2024}
}

The AlphaGeometry checkpoints and vocabulary are made available under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You can find details at: https://creativecommons.org/licenses/by/4.0/legalcode