High-Performance Tensor Network Library for Accelerating Quantum Science

We will present and describe the NVIDIA tensor network library, a state-of-the-art library of composable primitives for GPU-accelerated quantum physics simulations.

Tensor network methods have been a promising foundation and an effective research tool to explore the basic properties of quantum systems. The availability of a fast and scalable library becomes vital for quantum physics developers, as well as for quantum hardware engineers.

Our library was created to accelerate and scale up simulations developed by the quantum science community by enabling them to utilize efficient scalable software building blocks optimized for NVIDIA GPU-based platforms. Our goal is to open the space for new discoveries and a greater understanding of certain physical systems.

The building blocks provide easy access to all functionalities needed by tensor network methods including approximate tensor network algorithms based on matrix product state, and other factorized tensor representations. The rich capabilities provided are conveniently made available via both Python and C application programming interfaces.

Quantum science developers that have adopted the cuQuantum SDK demonstrated significant acceleration, compared to CPU-only execution, for tensor network simulation methods.