Research projects

Fundamentals in fluid fragmentation

During my Ph.D. at MIT, I investigated fluid fragmentation, a process critical to understanding how pathogens transmit from host to host. In epidemiology, the size and speed of droplets generated by infected hosts determine how fast a disease spreads through a population. However, the mechanism of fluid fragmentation has long been treated as a “black box” due to its rapid, complex sub-dynamics and a lack of precise measurement tools.

To overcome these challenges, I integrated advanced computer vision techniques to extract precise physical quantities directly from observation. Leveraging this data, I developed physics-based models for each stage of the fragmentation process, including sheet evolution, rim destabilization, ligament growth, and secondary droplet formation.

Through this work, I discovered two universal physical laws governing fragmentation and established a unified theoretical framework that links these sub-dynamics to predict final droplet distributions. These models, which show excellent agreement with experiments, pave the way for a fundamental understanding of fragmentation in both environmental health and industrial applications.

Achievement: 6 first-author papers in Journal of Fluid Mechanics and 1 first-author paper in Physical Review Letters

Relevant publications: [1, 2, 3, 4, 5, 6, 7]

Uncovering the hidden mechanics of Antarctic ice shelves

Antarctic ice shelves play a critical role in slowing down the flow of ice into the ocean, acting as a buffer against global sea-level rise. However, predicting their stability is difficult because their fundamental mechanical properties—specifically their flow law and viscosity structure—cannot be directly measured on a continental scale.

To solve this, I developed a novel approach combining vast remote-sensing datasets with physics-informed deep learning. This allowed me to “invert” the observations and infer the hidden rheology of glacial ice across Antarctica.

My research revealed that ice rheology is far more complex than previously thought. I found a distinct contrast between compression zones near the grounding line, which follow standard power laws, and extension zones, where the ice exhibits strong anisotropic properties. By accounting for this anisotropy, I constructed high-precision viscosity maps that capture suture zones—critical structures that inhibit rift propagation but are often missing from current models. These findings, published in Science, provide essential inputs for the next generation of ice-sheet models used to predict future sea-level rise.

Achievement: Two papers publised as first and corresponding author in Science and Journal of Open Source Software

Relevant publications: [10, 11, 12]

Discovery of unstable singularities with machine precision

Whether fluids can develop singularities—points where velocity or gradients become infinite—remains one of the most significant unanswered questions in mathematics. While stable singularities have been identified numerically, the real challenge of the field (such as the Millennium Prize Navier-Stokes problem) lies in finding unstable singularities. These solutions are exceptionally elusive because even microscopic perturbations can divert the system from its blow-up trajectory.

To tackle this, I developed a pioneering computational framework that merges deep learning with high-precision numerical optimization. Unlike traditional methods, our approach uses a multi-stage training scheme combined with a Gauss-Newton optimizer to navigate the complex landscape of nonlinear PDEs.

This framework led to the first systematic discovery of new families of unstable singularities in various fluid equations. Crucially, our method achieves unprecedented accuracy—on the order of O(10e−13)—limited only by hardware round-off errors. This extreme precision is a game-changer: it meets the stringent requirements needed to construct rigorous computer-assisted proofs, offering a new pathway to finally resolving these century-old mathematical mysteries

Achievement: Two first-author papers publised in Physical Review Letters and Journal of Computational Physics. Another two first-author papers are under review and available on Arxiv

Relevant publications: [8, 9, 13, 14]