Yuan Wang

Parallel Computing

In the field of simulating polariton condensates, there's a strong emphasis on leveraging advanced numerical methods, underpinned by a specialized computational tool that utilizes the capabilities of graphics processing units (GPUs). The split-step Fourier method (SSFM) is one of the central technique used for these simulations, offering a streamlined approach to solving partial differential equations in parallel computing environments. This method is also notably efficient when integrated into machine learning frameworks, offering innovative solutions for complex equations. My focus lies in developing computational tools based on NVIDIA's CUDA architecture, using C++ and CUDA APIs. This expertise encompasses both CPU-based and GPU-based GPE solvers, with a particular emphasis on utilizing SSFM for efficient problem-solving in the realm of quantum fluid dynamics.

Neural Operator

A plethora of next-generation all-optical devices based on exciton-polaritons have been proposed in latest years, including prototypes of transistors, switches, analogue quantum simulators and others. However, for such systems consisting of multiple polariton condensates, it is still challenging to predict their properties in a fast and accurate manner. The condensate physics is conventionally described by polariton Gross-Pitaevskii equations (GPEs). While GPU-based solvers currently exist, we propose a significantly more efficient machine-learning-based Fourier neural operator approach to find the solution to the GPE coupled with exciton rate equations, trained on both numerical and experimental datasets. The proposed method predicts solutions almost three orders of magnitude faster than CUDA-based solvers in numerical studies, maintaining the high degree of accuracy.

Positional embedding in ML

• Graph Neural Networks (GNNs) are powerful tools for molecular property prediction, but their reliance on local message-passing mechanisms limits their ability to capture long-range dependencies in large graphs. We propose a physics-informed approach that enhances graph convolutional networks (GCNs) by incorporating positional embeddings derived from photonic dynamics simulation.
• Photonic and polaritonic systems offer a fast and efficient platform for accelerating ML through physics-based computing. We propose a polaritonic ML approach for solving graph-based data problems. We demonstrate how lattices of condensates can efficiently embed relational and topological information from point cloud datasets. This information is then incorporated into a pattern recognition workflow based on convolutional neural networks (CNNs), leading to significantly improved learning performance compared to physics-agnostic methods. Our study introduces a distinct way of using photonic systems as fast tools for feature engineering, while building on top of high-performing digital machine learning.

Exciton-polariton condensate

The AlGaAs-like quantum wells within high-quality microcavities, can generate exciton-polaritons (polaritons) in the strong coupling regime. At cryogenic temperatures, the polariton condensates can be formed through optically excited high-energy excitons. I focus on the theoretical development of the methods for enhancing and focusing these condensates, utilizing localized nonresonant asymmetric-shaped excitation. A significant aspect of my work involves increasing spatial coherence and optimizing interaction strength between polaritons, paving the way for the realization of all-optical transistors and contributing to large-scale polariton condensates networks.