arXiv:2609.20406
The effects of shot noise on the quantum computation of NMR spectra
Sebastian Walch, Keith R. Fratus, Jan-Michael Reiner, Igor Lesanovsky
Abstract
Recent advances in the field of quantum computing hardware motivate the search for applications which demonstrate so-called quantum advantage. One promising use case that has been identified is the simulation of quantum many-body systems. The computational resources required for performing such a simulation on a classical computer generally grow exponentially with the size of the system being modeled, which is ultimately due to the exponential growth of the Hilbert space in which the dynamics of such a system take place. While digital quantum computers natively evolve quantum states directly in such a Hilbert space, thus naively avoiding this problem, the result of such a computation is typically not obtained as a deterministic output. Rather, it requires projective measurements which are fundamentally affected by shot noise. Any desired expectation values must therefore be reconstructed from repeated measurements, making the number of those measurements a relevant computational resource, and thus an important consideration for any potential claims of quantum advantage. In this work, we study how this resource scales with system size (a scaling which itself depends on the desired accuracy of the final result), using the simulation of nuclear magnetic resonance (NMR) spectra as a test case. We study this scaling for both real-world molecules, as well as a class of model NMR Hamiltonians which allow for efficient large-scale simulations with one-dimensional, two-dimensional, and all-to-all interactions. We find that the required resources increase only weakly with molecular size, far below the exponential growth of the underlying Hilbert space. This result suggests that shot noise should not pose a fundamental barrier to achieving quantum advantage in the simulation of NMR systems, and perhaps for many-body systems more broadly.
AI Summary
The paper asks whether measurement shot noise can impose an exponential resource cost on quantum computation of NMR spectra. It studies spectra reconstructed by Fourier transforming infinite-temperature magnetization correlation functions, with accuracy measured by cosine similarity to exact reference spectra rather than by resolving every transition. The measurement protocol samples initial product states in the $x$ basis, evolves them, and measures transverse magnetization. Each correlation estimate averages products of initial and final magnetization eigenvalues. With $R$ shots per time point and measurement direction, a covariance calculation and second-order expansion of cosine similarity give the large-$R$ mean spectral infidelity $\mathcal I(R,L)\approx\mathcal A(L)/R$. The system-size dependence is investigated numerically using analytically evaluable Ising dynamics for $L=10$–$50$ spins, with one-dimensional, two-dimensional, and all-to-all couplings. Gaussian noise generated from exact single-shot variances models finite-shot estimates. Across linewidths of $1$ and $3\,\mathrm{Hz}$, the inferred $\mathcal A(L)$ grows much more slowly than exponentially over the tested range, with pronounced flattening for distance-decaying interactions. Exact-diagonalization checks of Heisenberg models up to $12$ spins and four molecular examples also show weak size dependence. The study excludes hardware noise, decoherence errors, and time-evolution approximation errors. Its large-system results concern a classically tractable Ising model; the small Heisenberg calculations do not establish asymptotic scaling. The global spectral-overlap criterion discounts unresolved fine structure. The paper also notes that common liquid-state NMR spectra are already accurately accessible to approximate classical methods, so these results do not establish quantum advantage.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.