Developing GPU-accelerated numerical algorithms and parallel computing frameworks to enable extreme-scale computational fluid dynamics (CFD) simulations, addressing fundamental scalability bottlenecks arising from communication overhead and sequential algorithmic structures, and designing highly optimized solvers—such as pipelined tridiagonal matrix algorithms (TDMA)—that overlap communication and computation to maximize parallel efficiency, thereby achieving near-ideal scalability on multi-GPU systems and significantly accelerating core components of flow solvers, including Poisson solvers and implicit schemes.

Developing advanced computational fluid dynamics (CFD) methodologies to accurately resolve complex, nonlinear, and multi-scale physical phenomena in fluid systems, including turbulence, shock waves, multiphase flows, heat transfer, and fluid–structure interactions, and employing high-fidelity numerical methods and large-scale simulations to capture intricate flow physics that are difficult to predict using conventional approaches, thereby enabling deeper physical understanding and improved design of complex engineering systems.

Developing quantum algorithm frameworks for computational fluid dynamics (CFD) by reformulating fluid governing equations into forms amenable to quantum computation, leveraging operator splitting, conservative finite difference schemes, and artificial compressibility-based approaches such as the entropically damped artificial compressibility (EDAC) method, and designing quantum-compatible formulations that preserve key mathematical structures of fluid dynamics while enabling efficient simulation of Navier–Stokes equations on quantum computers, thereby advancing next-generation approaches for high-fidelity fluid simulation beyond the limits of classical computation.

Developing data-driven modeling frameworks based on deep neural networks to capture the high-dimensional and nonlinear dynamics of complex physical systems, leveraging advanced architectures such as CNNs, RNNs, and transformer-based models, and applying these frameworks to enable fast, accurate, and scalable predictions for computational fluid dynamics (CFD), surrogate and reduced-order modeling, inverse problems, and uncertainty quantification, ultimately facilitating real-time analysis and intelligent design of complex engineering systems.

Developing deep reinforcement learning (DRL)-based multi-constraint, multi-objective (MCMO) optimization frameworks that directly learn optimal design strategies across high-dimensional, nonlinear condition spaces, eliminating the need for repeated optimization at individual operating conditions and enabling efficient discovery of Pareto-optimal solutions over a continuous condition space, and integrating these frameworks with computational fluid dynamics (CFD) to perform end-to-end optimization—including automated mesh generation and design parameter tuning—such that optimal solutions balancing accuracy, computational cost, and physical performance can be obtained in a single-shot, non-iterative manner even for complex flow phenomena such as boundary layers, shocks, and flow separation.

Developing autonomous, AI-driven design and optimization frameworks for turbomachinery systems—including compressors, turbines, pumps, and propulsion devices—by integrating high-fidelity computational fluid dynamics (CFD), multi-condition multi-objective reinforcement learning (MCMO-RL), and automated computer-aided engineering workflows, and enabling robust, high-performance designs to be discovered across continuously varying operating conditions and complex multiphysics constraints, while leveraging multi-agent and multi-fidelity strategies to efficiently explore high-dimensional design spaces and identify globally optimal configurations for real-world applications such as data-center compressors, reactor coolant pumps (RCP), and advanced gas turbine systems.