Quantum Sensing as an Optimization-Inference Stack
Abstract
Every data science pipeline begins with a measurement. When that measurement is made by a quantum mechanical system—exploiting superposition, entanglement, and coherence—the sensor can in principle surpass the classical shot-noise precision limit. Atomic clocks already underpin GPS; magnetometers inform epilepsy-surgery planning; single-photon LiDAR systems detect methane leaks at parts-per-million-meter (ppmm) path-integrated concentration from ranges up to about 200 meters. This article has two purposes. The first is to serve as a primer for the INFORMS data science community; the author argues that the two central design problems in quantum sensing—what state should the probe be in? and what measurement should be performed?—can often be formulated or bounded using semidefinite programs (SDPs), develops both design problems from first principles, and shows how the resulting bounds constrain every other quantity in the sensing pipeline. The second purpose is to map a research agenda connecting the quantum layer to familiar data science tools: adaptive bandits, reinforcement learning, inverse problems, stochastic programming, and ML surrogates. An end-to-end methane emissions case study traces the complete stack from photon physics through SDP design to operational crew dispatch, providing the paper’s running quantum-versus-classical benchmark.
History: Yu Ding served as the senior editor for this article.

