acetn 0.1.3

GPU-accelerated corner-transfer-matrix renormalization of infinite projected entangled-pair states.

Ace-TN: Simple and Efficient iPEPS Simulation

About

Ace-TN is an efficient and easy-to-use library for simulating infinite projected entangled-pair states (iPEPS) using the corner-transfer matrix renormalization group (CTMRG) method.

• GPU Acceleration: Supports GPU computations, offering a significant performance boost for larger and more complex iPEPS simulations.
• Multi-GPU Support: Can utilize multiple GPUs, allowing users to scale calculations to the largest possible system sizes.
• Flexible and Easy to Use: Easy and straightforward API, providing easy management of tensors and flexible model Hamiltonian construction.

See our preprint at https://arxiv.org/abs/2503.13900.

Getting Started

Installation

To install Ace-TN, use:

pip install acetn

Quickstart

After installation, a simple example calculation can be run by using python3 script.py for the following script.py:

from acetn.ipeps import Ipeps

# Set simulation options in the config
config = {
    "TN":{
        "nx": 2,
        "ny": 2,
        "dims": {"phys": 2, "bond": 2, "chi": 20},
    },
    "model":{
        "name": "heisenberg",
        "params": {"J": 1.0},
    },
}

# Initialize an iPEPS
ipeps = Ipeps(config)

# Perform imaginary-time evolution steps
ipeps.evolve(dtau=0.01, steps=100)

# Measure observables
measurements = ipeps.measure()

For more details on usage, see the project documentation or try some of the example scripts in the samples directory.

Multi-GPU Usage

Run a script script.py in multi-GPU mode using N processes and N GPUs with

torchrun --nproc_per_node=N script.py

Citing

If you found Ace-TN useful for your research, please cite:

@misc{AceTN_2025,
      title={Ace-TN: GPU-Accelerated Corner-Transfer-Matrix Renormalization of Infinite Projected Entangled-Pair States}, 
      author={Addison D. S. Richards and Erik S. Sørensen},
      year={2025},
      eprint={2503.13900},
      archivePrefix={arXiv},
      primaryClass={cond-mat.str-el},
      url={https://arxiv.org/abs/2503.13900}, 
}

License

Ace-TN is licensed under the Apache-2.0 License. See the LICENSE file for more details.