Graphical Interfaces and Visualization

• TensorTrace
  TensorTrace is an application that facilitates the design and implementation of tensor network algorithms. Among many other features, TensorTrace can automatically determine the optimal contraction sequence and cost scaling for each network, and generate code (in MATLAB, Python or Julia languages) for contracting the tensor networks numerically.

High-Level Tensor Software

(Listed in alphabetical order by project name.)

• Cyclops Tensor Framework (C++)
  This library provides automatic parallelization of operations on multidimensional (sparse) arrays. Cyclops supports general tensor sparsity, so it is possible to define graph algorithms with the use of sparse adjacency matrices

• Cytnx (Python/C++)
  Cytnx is a library designed for Quantum/classical Physics simulations. Most of Cytnx APIs share very similar interfaces as the most common and popular libraries: numpy/scipy/pytorch. Cytnx also supports multi-devices (CPUs/GPUs) directly on the base container level. For algorithms in physics, Cytnx provides powerful tools such as UniTensor, Network, Bond, Symmetry etc. These objects are built on top of Tensor objects.

• Fastor (C++)
  Fastor is a stack-based high performance tensor (multi-dimensional array) library written in modern C++ [C++11/14/17] with powerful in-built tensor algebraic functionalities (tensor contraction, permutation, reductions, special tensor groups etc).

• FunFact (Python)
  FunFact enables flexible and concise expressions of tensor algebra through a hybrid NumPy/Einstein notation-based syntax. A particular emphasis is on automating the design of matrix and tensor factorization models. It’s areas of applications include quantum circuit synthesis, tensor decomposition, and neural network compression. FunFact is GPU-ready and easily parallizable thanks to the modern linear algebra backends such as JAX/TensorFlow and PyTorch that it builds on top of.

• ITensor (Julia, C++)
  ITensor is a library for tensor networks where tensor indices carry extra information and matching tensor indices automatically contract. Also features MPS and MPO algorithms, such as DMRG, and quantum number block-sparse tensors.

• NCON (MATLAB)
  NCON (Network Contractor) is a MATLAB routine which can be used for computing a broad class of tensor network diagrams in an efficient and convenient way.

• QUIMB (Python)
  python library for contracting tensors and tensor networks, with support for matrix product states and algorithms, and advanced features such as determination of the optimal ordering of tensor contractions.[1]

• Scikit-TT (Python)
  Scikit-TT is a Python library for applying the tensor-train (TT) format to various numerical problems in order to reduce the memory consumption and computational cost compared to classical tensor approaches significantly.

• SuSMoST
  SuSMoST (Surface Science Modelling and Simulation Toolkit) is meant to build and study classical lattice models of adsorption, such as Langmuir, Ising, Potts, or hard-core models. SuSMoST automatically builds a tensor network representation of a lattice model from samples of adsorption complexes. Several versions of the TRG algorithm are implemented in SuSMoST. They can be used to compute the partition function and its derivatives for simulated lattice models.

• SyTen (C++)
  A symmetry-protected tensor networks toolkit in C++, with a set of standard matrix-product state, binary tree-tensor network states and infinite projected entangled pair state utilities included.

• TeNeS (C++)
  TeNeS is a package for calculating many-body quantum states on two-dimensional lattices based on iTPS (iPEPS). It can treat various spin and boson Hamiltonians through comprehensive input parameters, while we need a simple input format for predefined lattices and models. Furthermore, it supports MPI/OpenMP hybrid parallelization for supercomputers.

• TensorKit (Julia)
  A Julia package for large-scale tensor computations, with a hint of category theory. TensorKit.jl aims to be a generic package for working with tensors as they appear throughout the physical sciences. Currently, most effort is oriented towards tensors as they appear in the context of quantum many body physics and in particular the field of tensor networks. Such tensors often have large dimensions and take on a specific structure when symmetries are present. To deal with generic symmetries, we employ notations and concepts from category theory all the way down to the definition of a tensor. At the same time, TensorKit.jl focusses on computational efficiency and performance.

• TenPy (Python)
  Tensor Network Python (TeNPy) is a Python library for the simulation of quantum systems with tensor networks. The philosophy of this library is to get a new balance of a good readability and usability for new-comers, and at the same time powerful algorithms and fast development of new algorithms for experts.[2]

• TensorLab (MATLAB)
  Tensorlab is a MATLAB toolbox for rapid prototyping of tensor decompositions with structured factors. By mixing different types of decompositions and factor structures, a vast amount of factorizations can be computed.

• TensorLy (Python)
  A python library offering a high-level API for tensor methods and deep tensorized neural networks. TensorLy’s backend system allows users to perform computations with NumPy, MXNet, PyTorch, TensorFlow, and CuPy in order to be scalable on both CPU and GPU.[3]

• TensorOperations (Julia)
  Julia package for tensor contractions and related operations. Offers fast tensor operations using a convenient Einstein index notation, as well as optimizing pairwise contraction order, the “ncon” function for contracting groups of tensors or tensor networks, GPU support, and other features.

• TensorNetwork (Python)
  python library for easy and efficient contraction of tensor networks, supporting multiple powerful backends

• Tensor Toolbox (MATLAB)
  The Tensor Toolbox is a MATLAB library supporting multiple tensor types, including dense, sparse, and symmetric tensors as well as specially structured tensors, such as Tucker format, and others. Tensors can be manipulated using MATLAB’s object-oriented features.

• TensorToolbox.jl (Julia)
  Julia package for tensors. Includes functionality for dense tensors, and tensors in the Tucker, Kruskal (CP), Hierarchical Tucker, and Tensor Train formats. Follows the functionality of MATLAB Tensor toolbox and Hierarchical Tucker Toolbox. Additionally, it contains algorithms from the paper Recompression of Hadamard Products of Tensors in Tucker Format by D. Kressner and L. Periša.

• TiledArray (C++)
  TiledArray is a massively-parallel, block-sparse tensor framework written in C++. It is a scalable, block-sparse tensor framework for rapid composition of high-performance tensor arithmetic, appearing for example in many-body quantum mechanics. It allows users to compose tensor expressions of arbitrary complexity in native C++ code that closely resembles the standard mathematical notation. The framework is designed to scale from a single multicore computer to a massive distributed-memory multiprocessor.

• TorchMPS (Python)
  TorchMPS is a framework for working with matrix product state (also known as MPS or tensor train) models within Pytorch. Our MPS models are written as Pytorch Modules, and can simply be viewed as differentiable black boxes that are interchangeable with standard neural network layers. However, the rich structure of MPS’s allows for more interesting behavior…

• TT-Toolbox (MATLAB)
  MATLAB implementation of basic operations with tensors in TT-format, including TT tensor and TT matrix formats, fast rounding procedures, methods for solutions of linear systems and eigenvalue problems, and the TT-cross method.

• Uni10 (C++)
  Universal Tensor Library, an open-source C++ library designed for the development of tensor network algorithms. Provides a Network class to process and store the details of the graphical representations of the networks, and supports quantum number block-sparse tensors.

• Xerus (C++)
  The Xerus library is a general purpose library for numerical calculations with higher order tensors, Tensor-Train Decompositions / Matrix Product States and general Tensor Networks. The focus of development was the simple usability and adaptibility to any setting that requires higher order tensors or decompositions thereof.

High-Performance, Lower-Level Tensor Software and Tensor Backends

• CuTensor (C)
  CuTensor is a library for high-performance contraction of arbitrary tensors on NVIDIA graphics processing units (GPUs). It is built on top of hardware-optimized, permutation-free contraction algorithms and can achieve very high performance.

• Cyclops Tensor Framework (C++)
  This library provides automatic parallelization of operations on multidimensional (sparse) arrays. Cyclops supports general tensor sparsity, so it is possible to define graph algorithms with the use of sparse adjacency matrices

• HPTT (C/C++)
  High-performance tensor transpose (HPTT) is a library for permuting (transposing) tensor data including summing two tensors which may have a different ordering of indices. Performance is achieved by hardware and cache-friendly algorithms as well as multi-threading based on OpenMP.

• taco (C/C++)
  taco is a library for compiling dense and sparse linear and tensor algebra expressions. The expressions can range from simple kernels like SpMV to more complex kernels like MTTKRP, where the operands can be dense, sparse, or a mix of dense and sparse. taco automatically generates efficient compute kernels (loops) to evaluate these expressions.

• TBLIS (C/C++)
  TBLIS is a contraction library for arbitrary (real-valued) tensors which uses a sophisticated permutation-free algorithm, avoiding the permutation overhead that would be incurred by a more naive permute and matrix-multiply algorithm.

Other Resources

• Review article “The landscape of software for tensor computations” by Psarras et al.[4] surveying and partially classifying existing tensor software.

References

1. quimb: a python library for quantum information and many-body calculations, Johnnie Gray, Journal of Open Source Software 3, 819 (2018)
2. Efficient numerical simulations with Tensor Networks: Tensor Network Python (TeNPy), Johannes Hauschild, Frank Pollmann, SciPost Phys. Lect. Notes, 5 (2018), 1805.00055
3. TensorLy: Tensor Learning in Python, Jean Kossaifi, Yannis Panagakis, Anima Anandkumar, Maja Pantic, 1610.09555
4. The landscape of software for tensor computations, Christos Psarras, Lars Karlsson, Paolo Bientinesi (2021), 2103.13756

Edit This Page