A preprint maps the quantum Fisher information of analytically tractable many body states onto a classical Monte Carlo problem, replacing exact values with bounds at system sizes where direct calculation fails.
For decades, physicists trying to figure out how precisely a quantum sensor can measure a magnetic field, an electric potential, or the ticking of an atomic clock have run into the same wall: the math. The figure of merit, the quantum Fisher information (QFI), encodes the maximum precision a quantum state can carry about some parameter of interest. Computing it for a realistic, noisy many-body state usually requires a full spectral decomposition of the density matrix, an operation whose cost explodes exponentially with system size and gives out long before the sensor itself would.
A preprint posted to arXiv on 21 May 2026 by Francesco Musso and Sara Murciano proposes a way around that wall, not by computing the QFI exactly, but by computing a systematic lower bound on it, using only classical sampling. The trick, the paper argues, is to treat certain analytically known many-body wave functions as if their amplitudes defined a probability distribution. Then classical Monte Carlo machinery can take over and produce estimates at system sizes that defeat exact diagonalization.
The catch, which the paper is unusually careful about, is that the result is a bound, not the true QFI. Whether the bound is tight depends on the noise channel, the wave function, and the observable used to probe the state. For the three physically motivated noise channels the author studies — local dephasing, local amplitude damping, and global depolarizing — the bound behaves differently, and the paper walks through why.
To understand the contribution, it helps to know what QFI is and why anyone bothers to bound it. A quantum sensor's precision in estimating some parameter (a field strength, a phase, a frequency) is limited by how much information the quantum state carries about that parameter. QFI is the formal measure of that information, and the standard precision limit in estimation theory says the variance of any unbiased estimator of the parameter scales as one over the QFI. Higher QFI means tighter measurement. The trouble is that for a real sensor, the state is not pure. Decoherence, the loss of quantum coherence from interaction with the environment, degrades it, and the QFI of a mixed state is much harder to compute than the QFI of a pure one. For an L-particle system described by a generic mixed density matrix, you need to know the entire spectrum, which is an O(2^L) problem.
Musso and Murciano's move is to restrict attention to a specific family of analytically known many-body states: Jastrow-Gutzwiller wave functions, originally developed in condensed matter physics to capture strong electron correlations. These wave functions are defined by amplitudes over occupation-number configurations, descriptions of which quantum states each particle occupies, that factor in a controlled way. The authors show that for such states, expectation values relevant to QFI can be rewritten as classical expectation values over a probability distribution given by the squared amplitudes in the occupation-number basis. This is the door through which Markov-chain Monte Carlo (MCMC) enters. MCMC is a standard computational technique that generates samples from a probability distribution by stepping through configurations and accepting or rejecting moves based on their relative probabilities. Once the mapping is in place, the QFI lower bound can be estimated by sampling, as detailed in the paper's full text on arXiv.
The cost is still exponential in system size L, but the exponent is small. Specifically, the paper reports a scaling of e^{b L} with b around 0.5 to 0.6, which the authors characterize as a "slow" exponential. In practice, this pushes the method into a regime well past what exact diagonalization can reach on a desktop, while still falling short of polynomial time.
The paper also does something useful beyond delivering a number. It identifies which observables maximize the QFI for the studied states, and it compares two families of lower bounds, polynomial and Krylov-based (the Krylov approach being a sequence of nested subspaces used to approximate how an operator acts on a state), under each of the three noise channels. The Krylov bounds behave differently from the polynomial ones in regimes where the effective rank of the noisy density matrix is small, and the authors relate that to the structure of the parameter-encoding operator. This is not just a numerical convenience. The choice of observable and the choice of bound both matter for how close the lower bound sits to the true QFI.
A few scope notes. The result is for Jastrow-Gutzwiller wave functions specifically. The authors describe the framework as extending naturally to other analytically known wave functions, but no demonstrations are given for those. The cost is exponential, not polynomial, and the bounds are lower bounds, not exact values. The paper is a two-author preprint with no peer review or independent replication yet visible. Francesco Musso is affiliated with Université Paris-Saclay and Pasqal SAS; Sara Murciano is affiliated with Université Paris-Saclay.
The result is classical in a precise sense. The estimator runs on a classical computer and samples from a classical distribution. It does not displace quantum computing and does not claim a quantum advantage. Its value is making a previously intractable quantum calculation tractable on classical hardware, with a clear, well-characterized bound as the price of admission.
For sensor builders, the practical upshot is modest but real. If your sensor's state happens to be well described by a Jastrow-Gutzwiller wave function, you can now estimate a lower bound on its ultimate precision under three common noise models, at system sizes that were out of reach a year ago. That does not tell you the sensor will work, but it tells you the precision floor.
What to watch next: whether independent groups reproduce the bound behavior on the same wave-function family, whether the framework extends cleanly to other analytically known states, and whether the Krylov-based bounds prove useful in regimes where the polynomial ones stagnate.