Adaptive Genetic Algorithms for Pulse-Level Quantum Error Mitigation

TL;DR

This work tackles the challenge of noise in NISQ quantum computing by introducing an adaptive genetic algorithm that optimizes pulse-level control parameters in real time, without altering the underlying circuit structure. The method leverages pulse-level representations within QuTiP-QIP to maximize fidelity under realistic noise modeled by Lindblad dynamics, applying the approach to Deutsch-Jozsa and Grover's algorithms. Key contributions include adaptive parameter tuning, preservation of circuit design, and empirical validation showing substantial fidelity gains over baseline pulses across multiple qubit counts and run lengths. The results suggest that evolutionary pulse-level optimization can significantly enhance quantum algorithm performance on noisy hardware, offering a practical pathway to more robust NISQ computations while highlighting the need for scalable strategies and robust hyperparameter management.

Abstract

Noise remains a fundamental challenge in quantum computing, significantly affecting pulse fidelity and overall circuit performance. This paper introduces an adaptive algorithm for pulse-level quantum error mitigation, designed to enhance fidelity by dynamically responding to noise conditions without modifying circuit gates. By targeting pulse parameters directly, this method reduces the impact of various noise sources, improving algorithm resilience in quantum circuits. We show the latter by applying our protocol to Grover's and Deutsch-Jozsa algorithms. Experimental results show that this pulse-level strategy provides a flexible and efficient solution for increasing fidelity during the noisy execution of quantum circuits. Our work contributes to advancements in error mitigation techniques, essential for robust quantum computing.

Figures (11)

• Figure 1: Workflow of the Adaptive Pulse-Level Error Mitigation Algorithm.
• Figure 2: Deutsch--Jozsa circuit applied to 4 qubits. The Hadamard gate on $q_3$ is placed in a separate ancilla from the controlled operations.
• Figure 3: Condensed 4-qubit Grover circuit marking $\lvert 1111\rangle$. The oracle (columns 2--6) adds a phase flip on $\lvert 1111\rangle$, while the diffusion operator (columns 7--13) amplifies that marked state.
• Figure 4: Average fidelity (avg) and standard deviation (std) for the Deutsch-Jozsa algorithm over 250 generations of pulse-level genetic optimization. The optimization process required approximately 1 hour and 30 minutes to complete using 6 cores. See Table \ref{['tab:vm_specs']} for CPU specifications. See Table \ref{['tab:deutschjozsa_experiments']} experiment with 4 qubits for more details with an early stop around before generation number 80.
• Figure 5: Baseline pulse waveforms for Deutsch-Jozsa obtained for the best-performing individual across all simulations.