Uni10

Official Resources

• Homepage: https://uni10.gitlab.io/
• Repository: https://github.com/yingjerkao/uni10
• License: LGPL (Lesser GNU Public License)
• Developers: Ying-Jer Kao Group (National Taiwan University)

Overview

Uni10 (Universal Tensor Network Library) is an open-source C++ library designed to facilitate the development and implementation of tensor network algorithms. It balances high-performance C++ execution with ease of use, providing a "Network" class to manage complex tensor diagrams intuitively. Uni10 is particularly focused on enabling researchers to write readable code for algorithms like DMRG, PEPS, and MERA while benefiting from behind-the-scenes optimization and GPU acceleration.

Scientific domain: Quantum Many-Body Physics, Tensor Networks. Target user community: Researchers developing custom tensor network codes (C++ or Python).

Theoretical Methods

• Tensor Operations: Contraction, Permutation, Reshaping, Bond fusion.
• Decompositions: SVD, QR, Eigendecomposition, LQ.
• Algorithmic Support: Primitives for building DMRG, iTEBD, PEPS, MERA.
• Symmetries: Support for Abelian symmetries (U(1), Z2) via block-sparse tensors.

Capabilities (CRITICAL)

• Network Class: Object-oriented management of tensor networks; handles index bookkeeping automatically.
• Python Wrappers: pyUni10 allows rapid prototyping in Python with C++ speed.
• GPU Support: (Uni10 v2+) Support for CUDA acceleration.
• Fermions: Handling of fermionic swap gates and statistics.
• Optimized Contraction: Heuristic algorithms to determine optimal pairwise contraction orders.

Key Strengths

Usability vs Performance

• Bridges the gap between high-level interpreted code (Python) and bare-metal C++.
• Encapsulates index management, reducing manual bookkeeping errors.

2D Focus

• Specifically designed with PEPS and MERA in mind, capable of handling higher-order tensors.

Inputs & Outputs

• Input: C++/Python scripts constructing tensors and networks.
• Output: Contracted values, Optimized tensors, Thermodynamic properties.

Interfaces & Ecosystem

• Language: C++ (Core), Python (Bindings).
• Dependencies: BLAS, LAPACK.
• Documentation: API references and tutorials for DMRG/MERA.

Advanced Features

• Block-Sparse Tensors: Efficient storage and computation for symmetric systems.
• Network Optimization: Internal graph analysis to minimize contraction cost.

Performance Characteristics

• Speed: Comparable to ITensor in benchmarks; overhead is minimal.
• Optimization: "Behind-the-scenes" memory management.

Comparison with Other Codes

• vs ITensor: Very similar philosophy (intelligent indices, C++). ITensor (Julia) is now more popular; Uni10 remains a strong C++ option with Python bindings.
• vs TeNPy: TeNPy is pure Python (mostly) + C modules; Uni10 is pure C++ + Python bindings.
• vs TensorNetwork: Uni10 is a specialized physics library; TensorNetwork is a general ML/Physics library.

Application Areas

• Spin Chains: Heisenberg, Hubbard models (1D).
• 2D Systems: PEPS simulations of square lattices.
• Critical Systems: MERA simulations of quantum critical points.

Best Practices

• Prototyping: Use pyUni10 to verify algorithms on small systems before moving to C++ for production.
• Symmetries: Always use Uni10::UniTensor with symmetries enabled for significant speedups.

Verification & Sources

Primary sources:

1. Repository: https://github.com/yingjerkao/uni10
2. Paper: "Uni10: An open-source library for tensor network algorithms" (arXiv:1511.05436).

Confidence: VERIFIED - Published and maintained. Verification status: ✅ VERIFIED