Three Birds with One Stone: Improving Performance, Convergence, and System Throughput with Nest

TL;DR

NEST presents a fidelity-aware execution framework for variational quantum algorithms that leverages intra-device qubit fidelity heterogeneity to simultaneously improve solution quality, accelerate convergence, and boost system throughput. By employing an Inverted ReLU ESP schedule, a structured qubit walk remapping strategy, and multi-programming, NEST navigates the trade-offs between exploration and exploitation of the quantum hardware's noise landscape. Empirical results on real IBM superconducting devices and simulations show NEST converges faster and achieves lower energy gaps than BestMap and Qoncord, while delivering higher throughput and lower user costs under a realistic cost model. The work demonstrates that treating fidelity as a tunable resource—and co-locating multiple VQAs on the same processor—can yield practical, scalable benefits for near-term quantum computing.

Abstract

Variational quantum algorithms (VQAs) have the potential to demonstrate quantum utility on near-term quantum computers. However, these algorithms often get executed on the highest-fidelity qubits and computers to achieve the best performance, causing low system throughput. Recent efforts have shown that VQAs can be run on low-fidelity qubits initially and high-fidelity qubits later on to still achieve good performance. We take this effort forward and show that carefully varying the qubit fidelity map of the VQA over its execution using our technique, Nest, does not just (1) improve performance (i.e., help achieve close to optimal results), but also (2) lead to faster convergence. We also use Nest to co-locate multiple VQAs concurrently on the same computer, thus (3) increasing the system throughput, and therefore, balancing and optimizing three conflicting metrics simultaneously.

Figures (18)

• Figure 1: The hybrid quantum-classical execution workflow of a variational quantum algorithm (e.g., VQE kandala2017hardware). The quantum circuit is shown here with four qubits (horizontal lines). The circuit starts in the ground state, indicated by $\ket{0}$, then the VQA circuit is run, and once completed, the qubits are measured, shown by the meters at the end.
• Figure 2: When a variational quantum eigensolver program is run on a high-fidelity map (Map 1), it achieves a closer to optimal performance (closer to the actual minimum energy), than when it is run on a low-fidelity map (Map 2) tannu2019not. Note: the quantum chip is laid out in IBM's heavy-hex topology jin2024optimizing. In this chip picture, the circles are the qubits, and the lines connecting them reflect qubit connections.
• Figure 3: Qoncord 10764550 models individual computers as high-fidelity devices (the ones with more high-fidelity qubits) and low-fidelity devices (the ones with fewer high-fidelity qubits); however, an opportunity exists within a computer to generate high-fidelity and low-fidelity maps by exploiting qubit noise profile heterogeneity murali2019noise.
• Figure 4: The ESP schedules we explored and how they compare to the optimal map and Qoncord techniques.
• Figure 5: Performance of different ESP schedules on noisy simulations for the H2 molecule VQE circuit Utkarsh2023Chemistryarrazola2021differentiable; the Inverted ReLU schedule performs the best. The energy gap reflects the percentage difference between the ideal minimum and the achieved minimum cost objective (energy in the case of VQE) -- lower is better. See Sec. \ref{['sec:methodology']} for methodological and implementation details for this experiment.