FlowDA: Accurate, low-latency weather data assimilation via flow matching
Data assimilation (DA) remains a major computational hurdle for AI weather forecasting. We introduce FlowDA, a flow-matching framework that fine-tunes the Aurora foundation model to deliver rapid, robust analyses. Unlike existing generative methods that suffer from error accumulation, FlowDA excels even with extreme data scarcity and maintains stability over long horizons—paving the way for scalable, real-time weather prediction.
Flipping the script: The microscopic secret of the inverse Magnus effect
In the macro-world, a spinning baseball curves through the air because of the Magnus effect, a phenomenon driven by inertia. However, at the microscopic scale of colloids, inertia is negligible and the fluid flow is time-reversible. Under these conditions, the Magnus effect should—theoretically—vanish. Yet, recent experiments (Cao et al., Nat. Phys., 2023) involving polymer solutions revealed a surprise: an inverse Magnus effect where colloids migrate in the direction opposite to what we see in Newtonian fluids.
Our work reveals that the inverse Magnus effect arises from stress-gradient-induced polymer transport, a mechanism that becomes dominant at colloidal scales. Unlike traditional models that assume a uniform polymer distribution, we demonstrate that migration is driven by a self-induced viscosity dipole—characterized by polymer accumulation at the front and depletion at the rear of the particle. Our first-principles theory quantitatively captures experimental observations without fitting, underscoring the vital role of microstructural inhomogeneities in complex fluids.
A theoretical framework for rapid design of phase change material systems
Phase change materials (PCMs) show promising potential for managing heat in Singapore's tropical climate, offering efficient thermal energy storage solutions. We develop a physics-based theoretical framework that links key design parameters to PCM performance through stability analysis and an easily interpreted melting curve, showing how heat is absorbed and released. The theory is rigorously validated against published experimental/numerical data and our own simulations on diverse PCM settings. They vary in material species—ranging from pure PCM to nanoparticle-enhanced PCM, in geometries—including an inclined rectangular cavity and an annular tube, and in operational conditions, such as isothermal and constant power heating. The non-data-driven, pen-and-paper model distills complex thermo-hydrodynamics into a practical map for rapid design and optimization of PCM systems. It delivers crucial mechanistic insights often elusive to simulations, experiments, or AI approaches alone. We invite collaboration with PCM-focused experimentalists and innovators in Singapore and worldwide, offering theoretical support for their investigations.
AI enables robots to propel and perform chemotaxis like microorganisms
We demonstrate chemotactic navigation of a multi-link articulated microrobot using two-level hierarchical reinforcement learning (RL). The lower-level RL allows the robot—featuring either a chain or ring topology—to acquire topology-adapted swimming gaits: wave propagation characteristic of flagella or body oscillation akin to an ameboid. These topology-adapted strokes capture inherent, topology-imposed constraints while emulating their biological counterparts. This adaptation illustrates how RL reproduces, in bionics, the morpho-functional relationship observed in biology—morphology defines functional patterns. Progressively, our higher-level RL allows both bionic swimmers to accomplish chemotactic navigation in prototypical scenarios, eliminating the need for manual resets. It further supports partially observable RL, where the robots observe only local scalar cues rather than global/vectorial data, thereby facilitating minimal onboard navigation. Our work demonstrates that RL, guided by biological insights, can advance novel locomotion robophysics and exemplifies the synergy of natural inspiration and AI in tackling microrobotic manipulation challenges.
Learn to swim like a beating flagellum: Transverse wave
Learn to swim like an oscillating ameboid: Longitudinal wave
Multiple faces of phoretic active matter: from crystalline solids to active turbulence
We employ large-scale, agent-resolved simulations to demonstrate that modulating the activity of a wet phoretic medium alone can govern its solid-liquid-gas phase transitions and, subsequently, laminar-turbulent transitions in fluid phases, thereby shaping its emergent pattern. These two progressively emerging transitions, hitherto unreported, bring us closer to perceiving the parallels between active matter and traditional matter. Our work reproduces and reconciles seemingly conflicting experimental observations on chemically active systems, presenting a unified landscape of phoretic collective dynamics.
​
Q. Yang^, M. Jiang^, F. Picano and L. Zhu*, Nat. Commun., 15, 2874, 2024​, selectively featured in Editor's Highlights. ^: equal contribution.
Active Wigner crystal
Liquid phase
Gas phase
Active turbulence
Data-driven intelligent manipulation of micro-scale particles
We present a data-driven architecture for controlling particles in microfluidics based on hydrodynamic manipulation. It replaces the difficult-to-derive model by a generally trainable artificial neural network to describe the particle kinematics, and subsequently identifies the optimal operations to manipulate particles. We demonstrate various manipulations, including targeted assembly of particles and subsequent navigation of the assembled cluster, simultaneous path planning for multiple particles, and steering one particle through obstacles.
W. Fang^, T. Xiong^, O. S. Pak and L. Zhu*, Adv. Sci., 2205382, 2022​. ​^: equal contribution.
Mimicking biological self-oscillations via an elasto-electro-hydrodynamic instability
Flagella and cilia beat or wiggle in a self-oscillatory fashion (not periodically actuated). Instead, prior studies commonly required a periodic power source to generate the forced oscillation of artificial filaments. Here, we engineer self-oscillation of artificial structures using an elasto-electro-hydrodynamic instability. We show numerically that applying a steady uniform electric field can produce the self-oscillatory locomotion of a microrobot composed of a dielectric particle and an elastic filament (see the video below). Linear stability analysis is performed to delineate this instability and the underlying physics. Experiments are then performed to realize this concept.
The strategy was conceived numerically in
​L. Zhu and H.A. Stone, Phys. Rev.Fluids, 4, 061701, Rapid Communications, 2019,
analyzed theoretically in
and realized experimentally in
E. Han, L. Zhu, J.W.Shaevitz and H. A. Stone, Proc. Natl. Acad. Sci. U.S.A., 118 (29), 2021.
A particle-encapsulating droplet in a creeping shear flow: instability and bifurcations
To understand the behavior of composite fluid particles such as nucleated cells and double emulsions in flow, we study a finite-size particle encapsulated in a droplet under creeping shear flow as a model system. In addition to its concentric particle-droplet configuration, we numerically explore other eccentric and time-periodic equilibrium solutions, which emerge spontaneously via supercritical pitchfork and Hopf bifurcations.
L. Zhu* and F.Gallaire, Phys.Rev.Lett., 119, 064502, 2017.