Tensor Network Enhanced Dynamic Multiproduct Formulas

Niall F. Robertson, Bibek Pokharel, Bryce Fuller, Eric D. Switzer, Oles Shtanko, Mirko Amico, Adam Byrne, Andrea D’Urbano, Salome Hayes-Shuptar, Albert Akhriev, Nathan Keenan, Sergey Bravyi, Sergiy ZhukPRX 2025 · DOI

TL;DR

This paper introduces a novel hybrid algorithm that combines tensor networks and quantum computation to simulate quantum systems more accurately and efficiently than either method alone. By using tensor networks to calculate coefficients and quantum computers for expectation values within a multiproduct formula framework, they achieve reduced circuit depths and improved accuracy for simulating Hamiltonian dynamics.

What Problem Does This Paper Address?

Simulating the time evolution of quantum many-body systems is crucial for understanding phenomena in chemistry, material science, and high-energy physics. However, this task is computationally demanding. Classical methods, like those using Matrix Product States (MPS), struggle with the exponential growth of entanglement, limiting them to small systems or short evolution times. While quantum computers theoretically offer a path to efficient simulation, current noisy, intermediate-scale quantum (NISQ) devices are plagued by errors (Trotter errors from approximate time evolution, and hardware noise like decoherence and crosstalk). This paper tackles the challenge of improving the accuracy and feasibility of quantum simulations on NISQ devices by synergistically combining the strengths of classical tensor network methods with quantum computation.

Methods

The core of the proposed method is a dynamic multiproduct formula (MPF) approach enhanced by Matrix Product Operators (MPOs).

  • Multiproduct Formulas (MPFs): Instead of using a single, potentially deep, Trotter circuit to approximate time evolution, MPFs combine multiple shallower Trotter circuits. The key idea is to use a linear combination of these shallower circuits, with coefficients optimized to minimize the overall error.
  • Dynamic MPFs: The paper focuses on dynamic MPFs, where these coefficients are time-dependent. This allows for better error minimization compared to static MPFs.
  • Matrix Product Operators (MPOs): A crucial innovation is the use of MPOs to represent and manipulate the complex operators involved in calculating the MPF coefficients. This is more memory-efficient than storing full quantum states as MPS, especially for longer simulation times. The MPO approach allows for the calculation of overlaps (which determine the MPF coefficients) without explicitly storing the time-evolved quantum states.
  • AQCtensor Algorithm: To further reduce circuit depth, the authors integrate the AQCtensor algorithm. AQCtensor finds a short-depth quantum circuit that approximates a target quantum state (prepared classically up to a certain time $t_1$) with high fidelity. This optimized circuit is then used as a starting point for further time evolution using shallower Trotter circuits.
  • Hybrid Workflow: The overall workflow involves:
  • Using AQCtensor to prepare a "quasi-exact" state up to time $t_1$.
  • Using MPOs on a classical computer to calculate the coefficients for a dynamic MPF, which approximates the time evolution from $t_1$ to $t$.
  • Using a quantum computer to prepare and measure the individual Trotter circuits needed for the MPF.
  • Combining the quantum measurements with the classically computed coefficients to obtain the final expectation values.

Datasets

The paper does not introduce new datasets for training. Instead, it uses a well-studied 50-qubit spin chain model (Heisenberg model and a modified version) as a benchmark for evaluating their simulation algorithm. The "datasets" are essentially the time-evolution data generated by simulating this model.

Key Insights and Ideas

  • Synergy is Key: The most significant insight is that combining tensor networks (MPOs) and quantum computation is not just an alternative but can be superior to using either in isolation. Tensor networks handle the complexity of operator manipulation classically, while quantum computers perform the necessary measurements efficiently.
  • MPOs for Operator Efficiency: Representing operators as MPOs offers a significant memory advantage over MPS for storing quantum states, especially as simulation time increases. This is critical for enabling longer-time simulations.
  • Reduced Circuit Depth = Reduced Noise: By using shallower Trotter circuits within the MPF framework and employing AQCtensor, the overall circuit depth is reduced. This directly translates to less accumulated noise and higher fidelity measurements on NISQ devices.
  • "Trotter Test" and "MPF Test": The authors introduce useful heuristics (vertical lines in their plots) to identify crossover points where MPFs start to outperform single Trotter circuits, and where the MPF advantage is rigorously guaranteed. This provides practical guidance for applying the method.

Key Results

  • Improved Accuracy: The proposed MPO-MPF algorithm consistently shows lower error (measured by Frobenius norm of the density matrix) compared to purely classical MPS simulations and purely quantum Trotter simulations, even with significantly lower classical memory requirements (lower MPO bond dimension).
  • Quantum Hardware Validation: The algorithm was successfully implemented on IBM's ibm_torino and ibm_kyiv quantum processors. Experimental results demonstrated that the dynamic MPF approach yields more accurate expectation values than single Trotter circuits, especially at later times, and is closer to theoretical predictions.
  • Circuit Depth Reduction: The AQCtensor enhancement, combined with MPF, allows for simulating quantum dynamics with circuits that are significantly shallower than those required for comparable accuracy using standard Trotterization. This is crucial for NISQ hardware.
  • Scalability Potential: The error analysis suggests that the MPO approach scales more favorably with time than MPS for storing quantum states, indicating potential for simulating larger systems or longer times.

Limitations and Future Work

  • 1D Spin Chain Focus: The current demonstration is limited to a 1D spin chain. While the MPO formalism is general, extending the method to higher-dimensional systems might require more sophisticated tensor network approaches beyond standard MPOs.
  • Error Model Assumptions: The error analysis relies on certain assumptions about error propagation. While the method aims to mitigate Trotter errors, it still depends on the effectiveness of hardware-level error suppression and mitigation techniques (like Pauli twirling and dynamical decoupling) to handle noise.
  • "Trotter Test" and "MPF Test" Interpretation: While these tests provide valuable insights, the authors note that the MPF can still be beneficial even outside the strictly guaranteed time regimes.
  • Future Work: The authors suggest exploring the application of this hybrid approach to more complex models, including higher dimensions. They also highlight the need for further numerical work to assess the practical performance in 2D and beyond. The integration with error mitigation techniques is also an ongoing area of research.

Summary updated

Abstract

Tensor networks and quantum computation are two of the most powerful tools for the simulation of quantum many-body systems. Rather than viewing them as competing approaches, here we consider how these two methods can work in tandem. We introduce a novel algorithm that combines tensor networks and quantum computation to produce results that are more accurate than what could be achieved by either method used in isolation. Our algorithm is based on multiproduct formulas (MPFs)—a technique that linearly combines Trotter product formulas to reduce algorithmic error. It uses a quantum computer to calculate the expectation values and tensor networks to calculate the coefficients used in the linear combination. We present a detailed error analysis of the algorithm and demonstrate the full workflow on a one-dimensional quantum simulation problem on <a:math xmlns:a="http://www.w3.org/1998/Math/MathML" display="inline"><a:mn>50</a:mn></a:math> qubits using two IBM quantum computers, and .
BibTeX
@inproceedings{robertson2025tensor,
  author = {Niall F. Robertson and Bibek Pokharel and Bryce Fuller and Eric D. Switzer and Oles Shtanko and Mirko Amico and Adam Byrne and Andrea D’Urbano and Salome Hayes-Shuptar and Albert Akhriev and Nathan Keenan and Sergey Bravyi and Sergiy Zhuk},
  title = {{Tensor Network Enhanced Dynamic Multiproduct Formulas}},
  booktitle = {PRX},
  year = {2025},
  doi = {10.1103/8bzc-dlgt},
}
Figure