Co-Designed Adaptive Quantum State Preparation Protocols

TL;DR

The paper addresses the hardware-software mismatch in near-term quantum computing by introducing Co-ADAPT-VQE, a hardware-aware extension of ADAPT-VQE that incorporates device connectivity, error rates, and gate costs into the ansatz-building process. By penalizing complex or poorly implemented excitations during operator selection, Co-ADAPT-VQE produces shallower, more noise-resilient state-preparation circuits for molecular ground states, demonstrated notably on linear nearest-neighbor architectures. Across H6 configurations, the method yields up to 97% reductions in two-qubit gate counts compared to hardware-agnostic baselines, with even larger benefits for strongly correlated systems, and shows meaningful improvements for ATA-connectivity as well. The framework is general and adaptable to various hardware setups and cost metrics, suggesting broad potential for enabling larger, more accurate quantum simulations on near-term devices.

Abstract

We propose a co-designed variant of ADAPT-VQE (Co-ADAPT-VQE) where the quantum hardware is taken into account in the construction of the ansatz. This framework can be readily used to optimize state preparation circuits for any device, addressing shortcomings such as limited connectivity, short coherence times, and variable gate errors. We exemplify the impact of Co-ADAPT-VQE by creating state preparation circuits for devices with linear nearest-neighbor (LNN) connectivity. We show a reduction of the CNOT count of the final circuits by up to 97% for 12-14 qubit systems, with the impact being greater for larger and more strongly correlated systems. Surprisingly, the circuits created by Co-ADAPT-VQE provide an over 70% CNOT count reduction with respect to the original ADAPT-VQE in all-to-all connectivity, despite being restricted to LNN qubit interactions.

Figures (21)

• Figure 1: Flowchart for the VQE algorithm.
• Figure 2: Flowchart for the ADAPT-VQE algorithm.
• Figure 3: Circuit implementation of the unitary $U_{\alpha_1\rightarrow\alpha_2}^{(QE)}(\theta)$, generated by qubit excitation $T_{\alpha_1\rightarrow\alpha_2}^{(QE)}$. The same circuit can be used for $U_{\beta_1\rightarrow\beta_2}^{(QE)}$. The circuit for $U_{\alpha_2\rightarrow\alpha_1}^{(QE)}$ (or $U_{\beta_2\rightarrow\beta_1}^{(QE)}$) is identical, but with the rotation angle flipped.
• Figure 4: Circuit implementation of the unitary $U_{\alpha_1\rightarrow\alpha_2}^{(FE)}(\theta)$, generated by qubit excitation $T_{\alpha_1\rightarrow\alpha_2}^{(FE)}$. The same circuit can be used for $U_{\beta_1\rightarrow\beta_2}^{(FE)}$. The circuit for $U_{\alpha_2\rightarrow\alpha_1}^{(FE)}$ (or $U_{\beta_2\rightarrow\beta_1}^{(FE)}$) is identical, but with the rotation angle flipped. The vertical dots indicate that there is a ladder of CNOT gates between qubits $q_{{z}_1}$ to $q_{|{\cal Z}_{JW}|}$.
• Figure 5: Circuit implementation of the unitary $U_{\alpha_1\beta_1\rightarrow\alpha_2\beta_2}^{(QE)}(\theta)$, generated by qubit excitation $T_{\alpha_1\beta_1\rightarrow\alpha_2\beta_2}^{(QE)}$. The circuit for $U_{\alpha_2\beta_2\rightarrow\alpha_1\beta_1}^{(QE)}$ is identical, but with the rotation angle flipped.