Error and Resource Estimates of Variational Quantum Algorithms for Solving Differential Equations Based on Runge-Kutta Methods

A focus of recent research in quantum computing has been on developing quantum algorithms for differential equations solving using variational methods on near-term quantum devices. A promising approach involves variational algorithms, which combine classical Runge-Kutta methods with quantum computations. However, a rigorous error analysis, essential for assessing real-world feasibility, has so far been lacking. In this paper, we provide an extensive analysis of error sources and determine the resource requirements needed to achieve specific target errors. In particular, we derive analytical error and resource estimates for scenarios with and without shot noise, examining shot noise in quantum measurements and truncation errors in Runge-Kutta methods. Our analysis does not take into account representation errors and hardware noise, as these are specific to the instance and the used device. We evaluate the implications of our results by applying them to two scenarios: classically solving a $1$D ordinary differential equation and solving an option pricing linear partial differential equation with the variational algorithm, showing that the most resource-efficient methods are of order 4 and 2, respectively. This work provides a framework for optimizing quantum resources when applying Runge-Kutta methods, enhancing their efficiency and accuracy in both solving differential equations and simulating quantum systems.

💡 Research Summary

This paper provides a rigorous error and resource analysis for variational quantum algorithms (VQAs) that solve differential equations by coupling classical Runge‑Kutta (RK) integration schemes with quantum circuit evaluations. The authors focus on two dominant sources of error: the truncation error inherent to the chosen RK method and the statistical shot‑noise error arising from finite‑sample quantum measurements. By deriving explicit upper bounds for both contributions, they obtain a total error bound expressed as the trace distance between the exact solution and the variational output.

The analysis begins with a concise review of RK methods. For a p‑th order RK scheme with s stages, the local truncation error (LTE) at each time step is bounded by ‖ℓ_n‖ ≤ K Δτ^{p+1} M L_f^{p} + O(Δτ^{p+2}), where K, M, and L_f are constants depending on the specific RK coefficients, the magnitude of the vector field f, and its Lipschitz constants. This bound is independent of the norm used, and the authors later estimate these constants for the concrete examples considered.

The variational quantum algorithm under study solves linear differential equations of the form d y/dτ = –H y, where H is Hermitian. By mapping the problem to imaginary‑time quantum evolution, the time‑dependent parameters of a parametrized quantum circuit obey exactly the same ODE. Each RK stage therefore requires evaluating the vector field f(τ, y) via expectation‑value measurements on a quantum circuit. The statistical error of these measurements scales as O(1/√N_meas), where N_meas is the number of shots per measurement.

The authors treat the two error sources as additive: ε_total ≤ ε_RK + ε_shot. For a prescribed target error ε, they allocate half of the budget to each component, leading to constraints ε_RK ≤ ε/2 and ε_shot ≤ ε/2. The RK constraint determines the step size Δτ (and thus the number of steps N_τ = T/Δτ) given the order p, while the shot‑noise constraint determines the required number of measurements per function evaluation, N_meas ∝ (ε_shot)^{-2}. Because each RK stage requires s(p) circuit evaluations, the total number of quantum circuit executions is

C(p) = s(p) · N_τ(p) · N_meas(p).

Higher‑order RK methods reduce the number of steps N_τ and the truncation error but increase the number of stages s(p) and, more importantly, the required measurement precision (since Δτ becomes smaller, the shot‑noise term grows as Δτ^{-(p+1)}). Consequently, the total cost C(p) is not monotonic in p; an optimal order exists that balances these competing effects.

To validate the theory, the paper presents two case studies. The first is a simple one‑dimensional ordinary differential equation. By numerically evaluating the derived bounds, the authors find that a fourth‑order RK scheme (p = 4) minimizes C(p). The second case is the Black‑Scholes partial differential equation for option pricing, discretized into a linear ODE of the form above. Here, a second‑order RK method (p = 2) yields the lowest total number of circuit evaluations. In both examples, the analytical upper bounds are deliberately conservative, and actual simulation errors are observed to be smaller, indicating that practical implementations may require fewer resources than the worst‑case estimates suggest.

The discussion emphasizes that the presented framework isolates the purely numerical aspects of the algorithm—truncation and shot‑noise—while deliberately excluding representation errors, gate infidelities, and other hardware‑specific noise sources. These omitted factors must be addressed in a full NISQ‑scale deployment, possibly through error mitigation or hardware‑aware ansatz design. Nonetheless, the paper’s contribution lies in providing the first comprehensive, analytically grounded resource model that integrates classical RK error analysis with quantum measurement statistics. This model enables practitioners to select the most resource‑efficient RK order for a given problem, potentially reducing the total runtime and measurement overhead dramatically.

Future directions suggested include extending the analysis to non‑linear PDEs, handling non‑Hermitian operators via embedding or alternative variational formulations, and incorporating realistic hardware noise models into the error propagation framework. By doing so, the community can move toward more accurate predictions of quantum advantage for differential‑equation solving on near‑term quantum devices.