Superiority of Krylov shadow tomography in estimating quantum Fisher information: From bounds to exactness

Estimating the quantum Fisher information (QFI) is a crucial yet challenging task with widespread applications across quantum science and technologies. The recently proposed Krylov shadow tomography (KST) opens a new avenue for this task by introducing a series of Krylov bounds on the QFI. In this work, we address the practical applicability of the KST, unveiling that the Krylov bounds of low orders already enable efficient and accurate estimation of the QFI. We show that the Krylov bounds converge to the QFI exponentially fast with increasing order and can surpass the state-of-the-art polynomial lower bounds known to date. Moreover, we show that certain low-order Krylov bound can already match the QFI exactly for low-rank states prevalent in practical settings. Such exact match is beyond the reach of polynomial lower bounds proposed previously. These theoretical findings, solidified by extensive numerical simulations, demonstrate practical advantages over existing polynomial approaches, holding promise for fully unlocking the effectiveness of QFI-based applications.

💡 Research Summary

The paper addresses the long‑standing challenge of efficiently estimating the quantum Fisher information (QFI), a central quantity in quantum metrology, entanglement detection, quantum algorithms, and quantum machine learning. Because the QFI is a highly nonlinear functional of the density matrix ρ and the generator H, direct estimation is infeasible for large‑scale quantum systems. Existing approaches therefore focus on estimating polynomial lower bounds (e.g., Legendre‑transform bounds, sub‑QFI bounds, Taylor‑expansion bounds). While these bounds are experimentally accessible, they inevitably leave a systematic gap to the true QFI that cannot be reduced by increasing the number of measurements.

The authors propose a fundamentally different framework called Krylov Shadow Tomography (KST). KST combines the Krylov subspace technique from numerical linear algebra with the shadow‑tomography protocol for estimating expectation values of polynomial functions of ρ. Starting from the commutator i