Complex systems, from engineered devices to natural environments involve multiple interacting physical processes across a range of spatial and temporal scales. These multi-physics phenomena, governed by partial differential equations (PDEs) and often involving phase transitions, can be highly sensitive to small variations in their components. I develop hybrid, physics-informed modeling frameworks to numerically simulate such systems. My work spans from cavitating and turbulent multi-phase flows in engineering applications to sea-ice dynamics in Earth system models. By coupling data-driven surrogates (such as Genetic Algorithms or Fourier Neural Operators) with traditional PDE-driven solvers, I build scalable and interpretable tools that accelerate simulation, enable uncertainty quantification, and bridge the gap between simulations and experiments or observations.
Accurately predicting the dynamics of a multi-physics system spanning wide temporal and spatial scales like, the Earth system requires both computationally efficient and differentiable models. My work in this area follows two complementary approaches.
First, I use Fourier Neural Operators (FNOs) as surrogate models for components of the Earth system, such as sea-ice dynamics. These surrogates capture complex, nonlinear spatiotemporal patterns directly from data, while being orders of magnitude faster than traditional solvers. This makes them especially valuable for uncertainty quantification. In addition, these surrogates can infer solutions for previously unseen conditions, effectively serving as emulators for multi-physics solver components.
Second, I work on enabling next-generation climate and ocean models in Julia like oceananigans.jl through differentiable programming and compiler technology. By leveraging automatic differentiation (AD) and performance-oriented compiler infrastructure, we are developing models that remain faithful to physical principles while being natively differentiable. This opens the door to integrating gradient-based optimization into global climate simulations, bridging the gap between numerical modeling, machine learning, and high-performance computing. This effort is being carried out in collaboration with researchers at MIT, UIUC and Argonne National Labs and supported by the National Science Foundation (NSF).
Together, these projects highlight a unified trajectory: building the foundation for hybrid physics–ML Earth system models that integrate physical knowledge with data-driven approaches while advancing computational efficiency.
Accurately modeling the interplay between cavitation and turbulence remains a major challenge. My work addresses this gap by systematically evaluating turbulence models for unsteady cavitating flows in nozzles and Venturi geometries.
Using Unsteady Reynolds-Averaged Navier Stokes (URANS) models, Detached Eddy Simulation (DES), and other hybrid URANS approaches, I investigate the ability of turbulence models to capture the characteristic cycle of cavity growth, shedding, and collapse. Results show that while standard turbulence models can reproduce global features such as cavity shedding and re-entrant jet dynamics, they consistently fall short in predicting local turbulence quantities like Reynolds stresses and turbulent kinetic energy. High-fidelity DES and 3D simulations reveal the importance of cavitation–vortex interactions, especially the role of velocity gradients and vorticity dynamics in shaping cavity morphology. By comparing numerical predictions against experimental measurements, these studies expose the bottleneck between turbulence model assumptions and the physics of cavitating flows, providing a reference for improving turbulence closures. I have further extended this work with uncertainty quantification techniques to assess model robustness and guide the development of more predictive, physics-informed turbulence models.
Apte, D., Ge, M., & Coutier-Delgosha, O. (2023). Numerical investigation of a cavitating nozzle for jetting and rock erosion based on different turbulence models. Geoenergy Science and Engineering, 231, 212300.
Apte, D., Ge, M., Zhang, G., & Coutier-Delgosha, O. (2024). Numerical investigation of three-dimensional effects of hydrodynamic cavitation in a Venturi tube. Ultrasonics Sonochemistry, 111, 107122.
Apte, D., Ge, M., & Coutier-Delgosha, O. (2021). Comparison of Reynolds shear stress methods for RANS turbulence modelling of a cloud cavitation in a venturi. In APS Division of Fluid Dynamics Meeting Abstracts (pp. P28-008).
Apte, D., Ge, M., & Coutier-Delgosha, O. Simulating Cloud Cavitation Using Detached Eddy Simulation and Other Hybrid Turbulence Models. In 8th International Conference on Multiphase Flow and Heat Transfer
Cavitation can severely impact hydraulic machinery like pumps and turbines, making it critical to identifying the mechanisms that drive its periodic behavior.
In collaboration with Aswin Gnanaskandan's team at Worcester Polytechnic Institute, I study cavitating flows using numerical simulations. We investigate cavitation over a wedge across a range of cavitation numbers using a combination of an in-house compressible solver and OpenFOAM. This cross-platform study assesses how different cavitation and turbulence models capture the sub-stages of a cavitating cycle and local cavity dynamics. Key findings show that while the models reproduce cavity behavior accurately and standard k-omega SST model is able to capture unsteady shedding without any empirical corrections, the DDES models is able to predict the void fraction accurately predictions at higher cavitation numbers, in a sharp contrast to URANS models.
In a previous experimental campaign at Virginia Tech, we measure liquid and vapor velocity fields in a small-scale Venturi section using laser-induced fluorescent particle image velocimetry. These measurements reveal significant slip velocities between the phases, challenging the common assumption of a homogeneous liquid–vapor mixture in computational models. Phase-locked analysis shows the slip occurs near the upper interface in upstream regions and near the bottom wall in the closure region due to re-entrant jets, providing critical reference data for improving multiphase flow models.
Ge, M., Apte, D., Wang, C., Zhang, G., Zhang, X., & Coutier-Delgosha, O. (2024). Slip velocity and field information of two-phase cavitating flows. Physics of Fluids, 36(9).
Apte, D., Lavari, M., Vaca-Revelo, D., Gnanaskandan, A., & Coutier-Delgosha, O. (2024, November). Simulation of cavitation over a wedge using various cavitation and turbulence models. In APS Division of Fluid Dynamics Meeting Abstracts (pp. X20-012).
Ge, M., Zhang, G., Apte, D., & Coutier-Delgosha, O. (2021). Analysis of the slip velocity between the two phases in a high-speed cavitating flow. In APS Division of Fluid Dynamics Meeting Abstracts (pp. M29-003).
Ge, M., Zhang, G., Apte, D., & Coutier-Delgosha, O. (2021). The Intensity and Topology Transition of Sheet/Cloud Cavitation at Elevated Temperatures. In 11 th International Symposium on Cavitation. Daejon, ROK