cuDensityMat
Type: Technology Tags: NVIDIA, CUDA, cuQuantum, quantum dynamics, density matrix, MPS, scientific computing Related: cuQuantum, cuTensorNet, cuStateVec, cuPauliProp, cuStabilizer, cuQuantum-Appliance, CUDA-Q, CUDA-QX, NVIDIA-Quantum, cuTENSOR, NCCL Sources: https://docs.nvidia.com/cuda/cuquantum/latest/cudensitymat/index.html, https://docs.nvidia.com/cuda/cuquantum/latest/ Last Updated: 2026-04-29
Summary
cuDensityMat is the cuQuantum library for accelerating analog quantum dynamics solvers. It provides APIs for defining pure or mixed quantum states in tensor-product spaces, defining many-body operators and superoperators, computing operator actions, propagating quantum states in time, computing expectation values, and evaluating operator eigenspectra on NVIDIA GPUs.
Detail
Purpose
Analog quantum dynamics workloads need to represent quantum states, operators, and master equations beyond simple gate-by-gate circuit simulation. cuDensityMat gives quantum researchers a GPU-accelerated library for operator actions, time propagation, gradients, expectation values, eigenspectrum workflows, and distributed execution in quantum many-body dynamics.
Current scope
- Pure and mixed quantum states in arbitrary tensor-product spaces, including MPS-factorized states.
- Many-body operators and superoperators represented as sums of elementary tensor operators, full matrix operators, matrix product operators, and coefficient terms.
- Operator and superoperator action on quantum states.
- Vector-jacobian products and gradients for parameterized operator/superoperator actions.
- Master-equation right-hand-side construction and coupled ODE system workflows.
- Expectation values, state/operator properties, time propagation, and extreme eigenspectrum computation.
- Linux support on x86_64 and ARM64 CPUs with Turing, Ampere, Ada, Hopper, and Blackwell GPUs.
NVIDIA context
cuDensityMat is part of the current five-library cuQuantum SDK alongside cuStateVec, cuTensorNet, cuPauliProp, and cuStabilizer. It depends on cuTensorNet and cuTENSOR in current NVIDIA docs, so it belongs in the same graph as tensor-network simulation and CUDA-Q quantum application workflows.
Connections
- cuQuantum - parent SDK and documentation family.
- cuTensorNet - dependency and adjacent tensor-network simulation component.
- cuStateVec - circuit state-vector simulation component.
- cuPauliProp and cuStabilizer - adjacent current cuQuantum libraries.
- cuQuantum-Appliance - containerized distribution including cuDensityMat.
- CUDA-Q and CUDA-QX - higher-level programming and application-library layers for quantum-classical workflows.
- NVIDIA-Quantum - platform context for accelerated quantum computing.
- cuTENSOR - tensor algebra dependency listed in current cuDensityMat prerequisites.
- NCCL - distributed GPU communication layer relevant to multi-GPU/multi-node execution.