Multiagent reinforcement learning for efficient variational fast-forwarding quantum circuits
Phys. Rev. A 113, 012618 – Published 21 January, 2026
DOI: https://doi.org/10.1103/qrmm-z56n
Abstract
Quantum simulation plays a pivotal role in understanding complex quantum systems. Yet, its implementation on noisy intermediate-scale quantum devices remains limited by hardware constraints. Variational fast forwarding (VFF), a hybrid quantum-classical algorithm, aims to address this challenge by approximating the diagonalization of short-time evolution operators. However, the original design of VFF relies on parametrized quantum circuits and , which often incur excessive gate complexity, impeding practical deployment on near-term quantum hardware. Meanwhile, current quantum circuit synthesis or compilation algorithms are insufficient for VFF circuit design, as this task requires the simultaneous construction of circuits and , while ensuring that approximates the eigenvectors and approximates the corresponding eigenvalues and remains diagonal. To overcome this limitation, we propose an approach that jointly designs shallow and circuits using multiagent reinforcement learning (MARL). In this fully cooperative framework, two agents collaboratively learn to construct the circuits while sharing rewards to optimize overall circuit performance. Our results demonstrate that the MARL-generated circuits generally achieve lower gate counts and tend to enable more reliable and faster convergence to the specified accuracy target compared to those produced by the standard VFF. Furthermore, our results highlight that our method typically outperforms qfast in terms of quantum gate counts, although occasional exceptions are observed. Simulated experiments on benchmark Hamiltonians confirm that our approach can reduce resource consumption and be used in situations where hardware capabilities limit the depth of quantum circuits.
