Abstract:
This cumulative thesis is concerned with theoretical quantum metrology, the theory of measurement and estimation using quantum resources. Possible applications of quantum-enhanced sensors include the measurement of magnetic fields, gravitational wave detection, navigation, remote sensing, or the improvement of frequency standards.
Many proposals for quantum-enhanced sensors rely on the preparation of non-classical initial states and integrable dynamics. However, such non-classical states are generally difficult to prepare and to protect against decoherence. As an alternative, in this thesis, we propose so-called quantum-chaotic sensors which make use of classical initial states that are easy to prepare while quantum enhancements are applied to the dynamics. This approach is motivated by the insight that quantum chaos and quantum metrology are both characterized by the sensitivity to small changes of the dynamics. At the example of the quantum kicked top model, where nonlinear control pulses render the dynamics quantum-chaotic, we explore different dynamical regimes for quantum sensors. Further, we demonstrate that quantum chaos is able to alleviate the detrimental effects of decoherence. In particular, we present a proposal for a quantum-chaotic cesium-vapor magnetometer.
With the help of reinforcement learning, we further optimize timing and strength of the nonlinear control pulses for the kicked top model with superradiant damping. In this case, the optimized control policy is identified as a dynamical form of spin squeezing.
Another part of this thesis deals with Bayesian quantum estimation and, in particular, with the problem of experiment design heuristics. We train neural networks with a combination of supervised and reinforcement learning to become fast and strong experiment design heuristics. We demonstrate the versatility of this method using examples of single and multi-parameter estimation where the trained neural networks surpass the performance of well-established heuristics.
Finally, this thesis deals with a long-time outstanding conjecture in quantum
metrology: we prove this conjecture and find an expression for the maximal quantum Fisher information for any mixed initial state and any unitary dynamics, provide conditions for optimal state preparation and optimal control of the dynamics, and utilize these results to prove that Heisenberg scaling can be achieved even with thermal states of arbitrary (finite) temperature.