[Paper Review] Demonstration of a programmable optical lattice atom interferometer

Catie LeDesma, Kendall Mehling+6 more|arXiv (Cornell University)|May 28, 2023

Cold Atom Physics and Bose-Einstein Condensates4 citations

TL;DR

This paper demonstrates a machine-designed optical lattice atom interferometer that uses reinforcement learning and quantum optimal control to manipulate ultracold atoms in a 1D optical lattice for high-precision inertial sensing. The system achieves near-optimal sensitivity for the enclosed space-time area, with a round-trip time under 500 μs and a potential single-atom sensitivity of ~6×10⁻⁸g when scaled to longer transport times.

ABSTRACT

Performing interferometry in an optical lattice formed by standing waves of light offers potential advantages over its free-space equivalents since the atoms can be confined and manipulated by the optical potential. We demonstrate such an interferometer in a one dimensional lattice and show the ability to control the atoms by imaging and reconstructing the wavefunction at many stages during its cycle. An acceleration signal is applied and the resulting performance is seen to be close to the optimum possible for the time-space area enclosed according to quantum theory. Our methodology of machine design enables the sensor to be reconfigurable on the fly, and when scaled up, offers the potential to make state-of-the art inertial and gravitational sensors that will have a wide range of potential applications.

Motivation & Objective

• To develop a reconfigurable, compact atom interferometer using optical lattices for enhanced robustness in dynamic environments.
• To overcome limitations of free-space atom interferometers in size, weight, and environmental stability for spaceborne and low-orbit applications.
• To leverage machine learning and optimal control to design complex quantum control sequences for interferometric components like beamsplitters and mirrors.
• To demonstrate high-precision acceleration sensing by measuring momentum state populations after recombination, enabling real-time performance tuning.
• To quantify performance against quantum limits and explore scalability toward shot-noise-limited operation and beyond via entanglement.

Proposed method

• Employed reinforcement learning (RL) and the PRojection Operator Newton method for Trajectory Optimization (PRONTO) to design time-dependent lattice shaking functions that manipulate atomic wavefunctions.
• Used a 1D optical lattice formed by counterpropagating 852 nm laser beams with phase and intensity controlled via acousto-optic modulators (AOMs), creating a periodic potential $ V(r,t) = V_L(r_\perp)\cos(2k\cdot r + \phi(t)) + V_D(r) $.
• Performed all-optical evaporation in a crossed 1064 nm dipole trap to produce $ 2 \times 10^4 $ Bose-Einstein condensates before loading into the lattice.
• Measured output populations in momentum states $ p \in \{-6\hbar k, \ldots, 6\hbar k\} $ after time-of-flight (TOF) imaging with a 780 nm probe laser.
• Calibrated the interferometer using maximum likelihood estimation based on Jensen-Shannon divergence between simulated and experimental momentum distributions.
• Quantified sensitivity using classical Fisher information $ I(a) = \sum_p \frac{1}{P(p|a)} \left( \frac{\partial P(p|a)}{\partial a} \right)^2 $, comparing to ideal free-space limits.

Experimental results

Research questions

• RQ1 Can machine learning and optimal control methods be used to design high-fidelity quantum control sequences for atom interferometry in an optical lattice?
• RQ2 To what extent does the performance of a lattice-based atom interferometer approach the quantum limit for a given space-time area?
• RQ3 How does the interferometer’s sensitivity scale with transport time, and what is the potential for achieving sub-10⁻⁷g sensitivity in practice?
• RQ4 Can the system be reconfigured in real time to tune dynamic range and sensitivity for varying measurement scenarios?
• RQ5 How accurately can the acceleration be estimated from the measured momentum state populations, and what is the role of asymmetry in direction discrimination?

Key findings

• The interferometer achieved experimental fidelities of 98.5% and 97.8% for 50/50 and 100/0 beamsplitters, respectively, as verified by time-of-flight images and momentum distributions.
• The system demonstrated direction-sensitive interference fringes with asymmetric momentum population distributions, enabling unambiguous acceleration direction discrimination.
• Fringe periodicity was tunable via transport time, allowing dynamic adjustment of sensitivity and bandwidth—longer times increased sensitivity but reduced bandwidth.
• The single-atom sensitivity was estimated at $ 6 \times 10^{-8}g $ when scaling transport time to 0.1 s, approaching the quantum limit for the enclosed space-time area.
• The Jensen-Shannon divergence between simulation and experiment showed high consistency, with calibrated points closely aligned to the diagonal, indicating reliable acceleration estimation.
• The device achieved a round-trip time of less than 500 μs, with atoms displaced only ~2.5 μm during 100 μs transport, highlighting compactness and potential for miniaturization.