Trapped atomic ions are used for quantum computing, atomic clocks, and tests of fundamental physics. Atomic vapor is used to obtain ions for trapping: bombard vapor with electrons or laser pulses to ionize atoms in it, then hold the ions in place with electromagnetic fields. Most existing ways to produce atomic vapor – ovens and laser ablation – are slow to turn on and consume significant power, limiting experimental up time and the size, weight, and power (SWaP) of trapped-ion experiments. We have designed a new method to make fast low-power ovens for calcium. The ovens are Ta discs that we coat with Ca and In by evaporation with traditional ovens. In serves as a protective layer to prevent oxidation of the Ca before use. The oven is later used by running current through the disc, outputting calcium vapor. The sub-2 mm diameter Ta disk has an extremely low heat capacity and is well thermally isolated from the environment, so that it can be heated up and start outputting Ca vapor in less than 1 s with less than 1 W of electrical power. My goal is to produce and test these ovens, finding the optimal thickness of the In layer in the process. I will present results from ongoing tests in which we measure the flux of atoms released from the oven by exciting an atomic transition with a laser and collecting the resulting atomic fluorescence. Viability of the oven production method will lower SWaP requirements for many experiments, expanding possibilities to test theories in promising but challenging environments like outer space.

The Epoch of Reionization marks the point in the history of the Universe when neutral atoms were reionized by the first stars and galaxies. Reionization spreads through the Universe in localized ionization bubbles before coalescing towards the end of reionization. Previous models generally predict a correlation between the density and ionization fields. Here, we present a suite of models that parameterize the correlation and simulate the reionization process from the underlying density data. The models use two methods for determining the evolution of reionization: mapping reionization directly from density data and propagating ionization from nearby ionized voxels. We investigate the best use of these methods through a comparison with 21cmFAST models. We then parameterize the density-ionization correlation and expand simulations to a variety of correlations. Finally, we use Markov chain Monte Carlo sampling to fit the parameters of the models with simulated HERA data. Specifically, we investigate how well future HERA data will constrain the density correlation parameter.

This project applies advanced machine learning techniques to two important societal challenges: improving Lyme disease diagnosis and addressing wrongful convictions in the U.S. legal system. For Lyme disease, which affects over 500,000 Americans annually, we analyzed patient data from the MyLymeData survey supported by LymeDisease.org using principal component analysis, nonnegative matrix factorization (NMF), and semi-supervised NMF (SSNMF). Our goal was to refine patient categorization into Neurological and Musculoskeletal groups based on patient symptoms. By evaluating symptom significance and diagnostic contexts, we improved the effectiveness of the classification. Our analysis also reveals that diagnostic circumstances are highly relevant to the chronic or acute patient classification, whereas both groups exhibit similar symptoms. This research paves the way for more accurate diagnostic frameworks. In the legal field, we worked to reduce biases in wrongful conviction cases by using AI to support The Innocence Project. Using selected murder cases from The National Registry of Exonerations and geographically matched non-exonerated cases, we developed a novel Convex Kernel SSNMF algorithm, extending Convex NMF and SSNMF to handle mixed-sign, nonnegative data. With polynomial kernels, our approach achieved higher classification accuracy on legal text embeddings and gene expression datasets compared to the benchmark algorithms. This dual-domain study shows how machine learning can help both healthcare and criminal justice.

Ultrafast Electron Diffraction (UED) is a cutting-edge technique that allows us to observe structural changes in materials on the femtosecond timescale. We use UED to investigate how the structure of Tantalum Disulfide (TaS₂) evolves in response to photoexcitation. By capturing these ultrafast dynamics, we aim to understand phase transitions and light-induced structural changes in quantum materials. An important part of this process involves preparing ultrathin TaS₂ samples which are needed to accurately image the samples. We will also discuss the challenges and methods involved in fabricating these high-quality samples suitable for UED.

The complete chip-scale integration of optical systems for ion trapping is vital to the development of compact quantum networks and portable precision measurement devices. The integration of stable, low-noise, narrowband lasers is a critical step in this process. Chip-scale stimulated Brillouin scattering (SBS) lasers are a promising alternative to bulkier, more commonly used external cavity diode lasers, which are used to decrease nonlinear noise in the lasers. Additionally, chip-scale coil resonators have the potential to provide compact laser stabilization by acting as a reference cavity. In this presentation, we demonstrate the use of a photonic integrated coil resonator and the SBS laser at the 88Sr+ trapped ion clock wavelength. We characterize the stability of the cavity and evaluate the losses of the both chips. Finally, we discuss the next steps for further chip-scale integration of the 88Sr+ trapped ion clock setup.