AI and quantum computing share one platform

Scientists have developed a new software platform, TensorCircuit-NG, to integrate quantum computing and simulation with artificial intelligence and high-performance computing. Shi-Xin Zhang of the Institute of Physics, Chinese Academy of Sciences, Yu-Qin Chen, and Weitang Li of the Graduate School of Engineering Physics of the Chinese Academy of Sciences collaborated with colleagues including Jiace Sun of the California Institute of Technology’s Department of Chemistry and Chemical Engineering and Pei-Lin Zheng of the China Mobile Research Institute for the Future to create the tensor-native programming paradigm. The framework leverages backends such as JAX, TensorFlow, and PyTorch to integrate circuits, tensor networks, and neural networks into a single differentiable computational graph. Importantly, TensorCircuit-NG addresses the limitations of existing simulators by providing scalable noise modeling and distributed computing strategies, enabling complex applications such as end-to-end quantum machine learning and differentiable optimization of tensor network states, and represents a significant advance in the field of quantum software development, with contributions from researchers at the Chinese Academy of Sciences and further collaborators.

TensorCircuit-NG accelerates progress by enabling researchers to simulate and optimize quantum systems at unprecedented scale and flexibility, potentially enabling practical applications in fields ranging from materials science to machine learning.

TensorCircuit-NG is a next-generation software platform that redefines the intersection of quantum physics, artificial intelligence, and high-performance computing. This framework directly addresses the exponential computational demands of complex quantum circuits through innovative distributed computing strategies such as automatic data parallelism and model-parallel tensor network slicing.

Validation on a GPU cluster demonstrates near-linear speedup of the distributed variational quantum algorithm, indicating a significant increase in computational efficiency. TensorCircuit-NG unlocks a streamlined pipeline for flagship applications, specifically end-to-end quantum machine learning for image recognition using the CIFAR-100 dataset and converting quantum states into neural networks via classical shadowing, a method to efficiently characterize quantum systems.

Additionally, this platform facilitates differentiable optimization of tensor network states, a critical step in tackling complex problems in many-body physics. The core of the innovation lies in the integration of disparate computational approaches, allowing researchers to seamlessly integrate quantum simulations with machine learning workflows, enabling the exploration of hybrid algorithms and modeling of physical systems at unprecedented scale.

By representing quantum computation as a differentiable graph, TensorCircuit-NG enables gradient-based optimization, the foundation of modern machine learning, to be applied directly to quantum circuit design and parameter tuning, promising to accelerate the discovery of new quantum algorithms and improve the performance of existing quantum algorithms. TensorCircuit-NG’s architecture is built around a two-layer design, unifying infrastructure and representation through a tensor-native philosophy.

This approach enables a flexible and interoperable ecosystem that supports a wide range of quantum systems, including quantum systems and fermion-Gaussian states. The platform’s advanced simulation engine, which includes analog, stabilizer, and approximate matrix product-state simulators, provides researchers with a versatile toolkit for exploring diverse quantum phenomena, alongside tools to model physical systems such as lattices, Hamiltonians, and time evolution, with comprehensive noise modeling and mitigation strategies.

GPU acceleration significantly reduces qubit and qubit Hamiltonian construction times

Constructing a sparse Hamiltonian matrix, specifically the transverse magnetic Ising model (TFIM), takes 0.059 seconds on a GPU using TensorCircuit-NG with a system size of 24 qubits. This is a significant performance improvement compared to traditional CPU-based methods, which require approximately 66.7 seconds for the same calculation. For a 22-qubit system, GPU-accelerated construction completes in 0.019 seconds, while CPU implementations using NumPy, QuSpin, and Quimb take 8.5 seconds, 4.3 seconds, and 10.9 seconds, respectively.

These timings, measured at complex128 precision on an NVIDIA H200 GPU, show orders of magnitude speedup compared to the standard library. This framework extends beyond qubits to efficiently simulate qubit systems and achieve seamless operation in local Hilbert spatial dimensions greater than 2. Simulations of quantum clock models using qutrits (d = 3) are performed using the dedicated QuditCircuit class, which generalizes gating rotations within arbitrary two-level subspaces.

This capability is demonstrated through ground state optimization using a just-in-time compiled and batched variational quantum eigensolver workflow. This implementation allows researchers to explore quantum systems with high-dimensional local Hilbert spaces while taking full advantage of the performance of the differentiable engine. TensorCircuit-NG includes a dedicated module for Gaussian states of fermions, allowing simulation of systems containing thousands of fermions with O(N3) computational complexity.

This module tracks two-body correlation matrices and reduces the amount of computation while maintaining the ability to compute multipoint correlations. Simulations of a 1D Kitaev chain reveal that the framework’s differentiable pipeline enables automatic discovery of phase boundaries and locates the critical chemical potential μc at 2t through gradient-rise maximization of entanglement entropy. The ability to compute gradients of macroscopic properties with respect to Hamiltonian parameters facilitates Hamiltonian learning and inverse engineering.

Lattice definition, visualization, and automatic Hamiltonian generation

TensorCircuit-NG takes a lattice-based approach to building and manipulating quantum systems. CustomizeLattice Class that allows creation of custom connectivity graphs. This class facilitates the dynamic addition or removal of sites and effectively models physical defects or impurities in the system while preserving the underlying coordinate information.

Take advantage of built-in visualization utilities to matplotlibrenders these grids, displays sites in physical coordinates, indicates nearest neighbor connections, and provides important visual feedback for debugging and validating geometric settings before computationally intensive simulations. This framework further automates Hamiltonian structures by: tc.templates.hamiltonians Use the module to generate operator lists directly from defined lattice objects.

This entire pipeline, from coordinate definitions to Hamiltonian structures, is designed to be end-to-end differentiable and “jitterable” and can be compiled to improve performance. This capability allows optimization of physical geometric parameters such as lattice constants to minimize ground state energy or target specific spectral properties.

In the example, optax A library for gradient-based optimization. To address common bottlenecks in many-body quantum simulations, TensorCircuit-NG includes: PauliStringSum2COO Function, JIT-compiled implementation for GPU acceleration.

This function constructs a sparse Hamiltonian matrix in a fully vectorized manner, computing the coordinate list indices and nonzero values ​​of all Pauli terms in parallel rather than sequentially. We see significant performance improvements when benchmarking against standard CPU-based libraries (NumPy, QuSpin, Quimb) on the NVIDIA H200 GPU. A sparse matrix with L = 24 is constructed in less than 60 milliseconds, compared to tens of seconds using traditional methods. This accelerated construction also features end-to-end differentiation and jitterability, further increasing the framework’s versatility.

Integrating quantum circuits, tensor networks, and machine learning for scalable simulations

The relentless pursuit of simulating quantum systems has long been hampered by computational costs. Quantum computers promise to overcome this problem, but they are still in their infancy and error-prone. TensorCircuit-NG represents a major advance in addressing this challenge by radically optimizing rather than circumventing classical limitations. This is not just a fast simulator. This is a fundamentally different approach that integrates the languages ​​of quantum circuits, tensor networks, and machine learning into a single differentiable framework.

Its impact extends beyond simply modeling large systems. Researchers have struggled for years to bridge the gap between theoretical quantum algorithms and real-world machine learning applications. TensorCircuit-NG provides a pathway to seamlessly integrate these fields, enabling end-to-end differentiable quantum machine learning, a critical step toward realizing the potential of quantum-enhanced artificial intelligence.

The framework’s ability to handle complex mechanics, including those found in many-body physics, opens the door to materials discovery and fundamental scientific research. However, relying on classic machine learning backends has inherent limitations. Although the speedup is significant, it is still limited by classical computational resources. Furthermore, the accuracy of approximate simulations remains a significant concern.

Although the platform incorporates advanced noise modeling, accurately capturing all the complexities of real-world quantum hardware is an ongoing battle. Future developments may focus on improving these approximations and exploring hybrid quantum-classical algorithms that leverage the strengths of both paradigms. The real test will be whether this unified framework can be extended to address the currently difficult problems for both classical and quantum computers, and whether it can truly accelerate the development of practical quantum technologies.