Machine learning estimation of optimal resolved sideband cooling strategies

Ground state preparation of trapped ions through resolved sideband cooling is an integral technique to trapped ion experiments and technologies. Despite its prevalence, the optimal cooling strategy given an initial ion temperature, a target temperature, and trap parameters is currently unknown. We present a methodic search for the fastest cooling strategies over a wide range of experimental parameters. By varying the laser pulse duration and number of pulses, and by including the possibility of higher-order sidebands, we determine potential candidates for optimal cooling strategies. Since the space of cooling strategies scales exponentially both in the number of pulses and the number of higher-order sidebands applied, brute-force computation proves impractical if not impossible. Instead, we numerically estimate optimal cooling strategies by employing a support-vector machine trained on Monte Carlo estimators. We find that traditional sequential pi pulse strategies are far from optimal, especially for initially "hot" ions and/or ions not deep within the Lamb-Dicke regime, and propose more efficient pulse sequences for achieving faster near-ground-state cooling.