2025-04-10 | | Total: 13
Kolmogorov-Arnold Networks (KANs) have gained attention as a promising alternative to traditional Multilayer Perceptrons (MLPs) for deep learning applications in computational physics, especially within the framework of physics-informed neural networks (PINNs). Physics-informed Kolmogorov-Arnold Networks (PIKANs) and their variants have been introduced and evaluated to solve inverse problems. However, similar to PINNs, current versions of PIKANs are limited to obtaining solutions for a single computational domain per training run; consequently, a new geometry requires retraining the model from scratch. Physics-informed PointNet (PIPN) was introduced to address this limitation for PINNs. In this work, we introduce physics-informed Kolmogorov-Arnold PointNet (PI-KAN-PointNet) to extend this capability to PIKANs. PI-KAN-PointNet enables the simultaneous solution of an inverse problem over multiple irregular geometries within a single training run, reducing computational costs. We construct KANs using Jacobi polynomials and investigate their performance by considering Jacobi polynomials of different degrees and types in terms of both computational cost and prediction accuracy. As a benchmark test case, we consider natural convection in a square enclosure with a cylinder, where the cylinder's shape varies across a dataset of 135 geometries. We compare the performance of PI-KAN-PointNet with that of PIPN (i.e., physics-informed PointNet with MLPs) and observe that, with approximately an equal number of trainable parameters and similar computational cost, PI-KAN-PointNet provides more accurate predictions. Finally, we explore the combination of KAN and MLP in constructing a physics-informed PointNet. Our findings indicate that a physics-informed PointNet model employing MLP layers as the encoder and KAN layers as the decoder represents the optimal configuration among all models investigated.
Accurate modeling of the complex dynamics of fluid flows is a fundamental challenge in computational physics and engineering. This study presents an innovative integration of High-Order Singular Value Decomposition (HOSVD) with Long Short-Term Memory (LSTM) architectures to address the complexities of reduced-order modeling (ROM) in fluid dynamics. HOSVD improves the dimensionality reduction process by preserving multidimensional structures, surpassing the limitations of Singular Value Decomposition (SVD). The methodology is tested across numerical and experimental data sets, including two- and three-dimensional (2D and 3D) cylinder wake flows, spanning both laminar and turbulent regimes. The emphasis is also on exploring how the depth and complexity of LSTM architectures contribute to improving predictive performance. Simpler architectures with a single dense layer effectively capture the periodic dynamics, demonstrating the network's ability to model non-linearities and chaotic dynamics. The addition of extra layers provides higher accuracy at minimal computational cost. These additional layers enable the network to expand its representational capacity, improving the prediction accuracy and reliability. The results demonstrate that HOSVD outperforms SVD in all tested scenarios, as evidenced by using different error metrics. Efficient mode truncation by HOSVD-based models enables the capture of complex temporal patterns, offering reliable predictions even in challenging, noise-influenced data sets. The findings underscore the adaptability and robustness of HOSVD-LSTM architectures, offering a scalable framework for modeling fluid dynamics.
Deng, Hani, and Ma [arXiv:2503.01800] claim to resolve Hilbert's Sixth Problem by deriving the Navier-Stokes-Fourier equations from Newtonian mechanics via an iterated limit: a Boltzmann-Grad limit (ε→0, N \varepsilon^{d-1} = \alpha fixed) yielding the Boltzmann equation, followed by a hydrodynamic limit (\alpha \to \infty) to obtain fluid dynamics. Though mathematically rigorous, their approach harbors two critical physical flaws. First, the vanishing volume fraction (N \varepsilon^d \to 0) confines the system to a dilute gas, incapable of embodying dense fluid properties even as \alpha scales, rendering the resulting equations a rescaled gas model rather than a true continuum. Second, the Boltzmann equation's reliance on molecular chaos collapses in fluid-like regimes, where recollisions and correlations invalidate its derivation from Newtonian dynamics. These inconsistencies expose a disconnect between the formalism and the physical essence of fluids, failing to capture emergent, density-driven phenomena central to Hilbert's vision. We contend that the Sixth Problem remains open, urging a rethink of classical kinetic theory's limits and the exploration of alternative frameworks to unify microscale mechanics with macroscale fluid behavior.
In recent years, research and development in nanoscale science and technology have grown significantly, with electrical transport playing a key role. A natural challenge for its description is to shed light on anomalous behaviours observed in a variety of low-dimensional systems. We use a synergistic combination of experimental and mathematical modelling to explore the transport properties of the electrical discharge observed within a micro-gap based sensor immersed in fluids with different insulating properties. Data from laboratory experiments are collected and used to inform and calibrate four mathematical models that comprise partial differential equations describing different kinds of transport, including anomalous diffusion: the Gaussian Model with Time Dependent Diffusion Coefficient, the Porous Medium Equation, the Kardar-Parisi-Zhang Equation and the Telegrapher Equation. Performance analysis of the models through data fitting reveals that the Gaussian Model with a Time-Dependent Diffusion Coefficient most effectively describes the observed phenomena. This model proves particularly valuable in characterizing the transport properties of electrical discharges when the micro-electrodes are immersed in a wide range of insulating as well as conductive fluids. Indeed, it can suitably reproduce a range of behaviours spanning from clogging to bursts, allowing accurate and quite general fluid classification. Finally, we apply the data-driven mathematical modeling approach to ethanol-water mixtures. The results show the model's potential for accurate prediction, making it a promising method for analyzing and classifying fluids with unknown insulating properties.
We are concerned with the low regularity of self-similar solutions of two-dimensional Riemann problems for the isentropic Euler system. We establish a general framework for the analysis of the local regularity of such solutions for a class of two-dimensional Riemann problems for the isentropic Euler system, which includes the regular shock reflection problem, the Prandtl reflection problem, the Lighthill diffraction problem, and the four-shock Riemann problem. We prove that it is not possible that both the density and the velocity are in H^1 in the subsonic domain for the self-similar solutions of these problems in general. This indicates that the self-similar solutions of the Riemann problems with shocks for the isentropic Euler system are of much more complicated structure than those for the Euler system for potential flow; in particular, the density and the velocity are not necessarily continuous in the subsonic domain. The proof is based on a regularization of the isentropic Euler system to derive the transport equation for the vorticity, a renormalization argument extended to the case of domains with boundary, and DiPerna-Lions-type commutator estimates.
In this paper, we showcase how flow obstruction by a deformable object can lead to symmetry breaking in curved domains subject to angular acceleration. Our analysis is motivated by the deflection of the cupula, a soft tissue located in the inner ear that is used to perceive rotational motion as part of the vestibular system. The cupula is understood to block the rotation-induced flow in a toroidal region with the flow-induced deformation of the cupula used by the brain to infer motion. By asymptotically solving the governing equations for this flow, we characterise regimes for which the sensory system is sensitive to either angular velocity or angular acceleration. Moreover, we show the fluid flow is not symmetric in the latter case. Finally, we extend our analysis of symmetry breaking to understand the formation of vortical flow in cavernous regions within channels. We discuss the implications of our results for the sensing of rotation by mammals.
This study investigates unsteady boundary layer phenomena in electrically conducting fluids subjected to static magnetic fields. Using a semi-explicit similarity transformation method, the momentum equation associated with the Stokes stream function is solved. The nonlinear closed analytical solutions for both stagnation flow and converging flow are derived. The results demonstrate that the boundary layer structure incorporates similar shock and solitary wave components which are promoted by Lorentz force. Under extreme magnetic fields, the flow exhibits sine and cosine wave patterns, which are motivated by the strong Lorentz force. An in-depth asymptotic analysis establishes the square root scaling laws that quantify the growth of friction and flux with increasing magnetic field strength. The boundary layer thickness scales inversely with the Hartmann number, a consequence of dominant Lorentz force, which differs from the conclusion of duct flow (Hunt 1965). These findings elucidate the physical mechanisms governing the nonlinear coupling between magnetic fields and the dynamics of the boundary layer.
Thermocapillary motion is widespread in both natural and engineering applications. A tiny drop of one liquid, suspended within another, may be set into motion aligned with an imposed thermal gradient, as influenced by thermocapillary action stemming from the gradients in interfacial tension due to the local variations in temperature. In real-world situations, however, such drops do not remain in isolation, as they interact with their neighboring entities, including other drops in proximity as well as a nearby solid boundary, setting up a complex interplay between the confinement-mediated interactions and the three-dimensional nature of the droplet dynamics. In this study, we present numerical solutions for the migration dynamics of a tightly confined drop couple, incorporating deformable interfaces, film flow, and Marangoni effects in the presence of dynamically evolving thermocapillary stresses induced by an imposed uniform temperature gradient. Unlike prior investigations, our work highlights the influence of the confinement towards orchestrating non-trivial features of drop migration, as dictated by an intricate coupling of the thermal and flow fields amidst the interferences of the domain boundaries. The study reveals that hydrodynamic interactions resulting from a juxtaposition of these influences deform the drops in a unique manner as compared to the characteristics evidenced by previously reported studies, causing a distortion of the local thermal fields around them. This, in turn, leads to changes in the local thermocapillary stress, affecting the local shear gradient in a manner that alters the local flow field in accordance with ensuring the interfacial stress balance.
Direct numerical simulations of a uniform flow past a fixed spherical droplet are performed to investigate the parameter range within which the axisymmetric flow becomes unstable due to an external flow bifurcation. The hydrodynamics is governed by three dimensionless numbers: the viscosity ratio, \mu^\ast, and the external and internal Reynolds numbers, \Rey^e and \Rey^i, respectively. The drop-to-fluid density ratio is related to these parameters as \rho^\ast=\mu^\ast \Rey^i/\Rey^e. This study focuses on highly viscous droplets with \mu^\ast \geq 5, where wake instability is driven by the vorticity flux transferred from the droplet surface into the surrounding fluid. By analysing the wake structure, we confirm that the onset of the external bifurcation is linked to the tilting of the azimuthal vorticity, \omega_\phi, in the wake and that the bifurcation occurs once the isocontours of \omega_\phi align nearly perpendicular to the symmetry axis. We propose an empirical criterion for predicting the onset of the external bifurcation, formulated in terms of the maximum vorticity on the external side of the droplet surface. This criterion is applicable for sufficiently high \Rey^i and holds over a wide range of \mu^\ast and \Rey^e. Additionally, we examine the bifurcation sequence for two specific external Reynolds numbers, \Rey^e=300 and \Rey^e=500, and show that, beyond a critical viscosity ratio, the axisymmetric wake first transitions to a steady planar-symmetric state before undergoing a secondary Hopf bifurcation. Finally, we highlight the influence of \Rey^i on external bifurcation and show that, at moderate \Rey^i, wake instability may set in at a lower vorticity threshold than predicted by our criterion. These findings provide new insights into the external flow bifurcation of viscous droplets.
We characterize the electrokinetic flow due to the transport of electrolytes embedded in nanochannels of varying cross-section with inhomogeneous slip on their walls, modeled as an effective slip length on the channel wall. We show that, within linear response and Debye-Huckel regime, the transport coefficients, and so the fluxes, can be significantly improved by the presence of a hydrophobic surface coating located at the narrowest section of the nanochannel. Our model indicates that the enhancement is larger when considering electric conductive walls in comparison to dielectric microchannel walls, and it is produced by a synergy between the entropic effects due to the geometry and the presence of the slip boundary layer. Our results show that a tailored hydrophobic coating design can be an effective strategy to improve transport properties in the broad areas of lab-on-a-chip, biophysics, and blue energy harvesting and energy conversion technologies.
This study presents a novel approach for enhancing Reynolds-averaged Navier-Stokes (RANS) turbulence modeling through the application of a Relative Importance Term Analysis (RITA) methodology to develop a new zonally-augmented k-\omega SST model. Traditional Linear Eddy Viscosity Models often struggle with separated flows. Our approach introduces a physics-based binary classifier that systematically identifies separated shear layers requiring correction by analyzing the relative magnitudes of terms in the turbulence kinetic energy equation. Using symbolic regression, we develop compact correction terms for Reynolds stress anisotropy and turbulent kinetic energy production. Trained on two-dimensional configurations, our model demonstrates significant improvements in predicting separation dynamics while maintaining baseline performance and fully attached flows. Generalization tests on Ahmed body and Faith Hill three-dimensional configurations confirm robust transferability, establishing an effective methodology for targeted enhancement of RANS predictions in separated flows.
Wall-bounded turbulence is characterized by coherent, worm-like structures such as hairpin vortices. The organization of these vortices near the wall can be modeled using the attached-eddy hypothesis, which effectively describes velocity statistics in the log-law region and energy-containing motions. However, the complex geometry and diverse scales of vortex structures in wall turbulence present significant challenges for developing and applying physics-based models. Here, we model wall turbulence as an ensemble of complex vortices, offering a systematic approach to constructing turbulence fields enriched with hierarchically organized hairpin vortex packets at a range of Reynolds numbers. The geometry and distribution of these vortex packets are calibrated to match physical observations and flow statistics. Our model successfully reproduces key features of wall turbulence, including mean velocity profiles, higher-order velocity fluctuation moments, near-wall streaks, and energy spectra that align with theoretical and numerical findings of turbulent channel flows at friction Reynolds numbers from 1,000 to 10,000. Furthermore, the constructed channel turbulence transitions rapidly to fully developed turbulence through direct numerical simulation, demonstrating its practicality for generating inlet or initial conditions in numerical simulations. This approach significantly reduces computational costs associated with turbulence development while providing a robust framework for testing and advancing turbulence models based on vortex structures.
While the lattice Boltzmann method (LBM) has proven robust in areas like general fluid dynamics, heat transfer, and multiphase modeling, its application to mass transfer has been limited. Current modeling strategies often oversimplify the complexities required for accurate and realistic mass transfer simulations in multicomponent miscible mixtures involving external forces. We propose a forcing approach within the explicit velocity-difference LBM framework to address these limitations. Our approach recovers the macroscopic mass conservation equations, the Navier-Stokes equation with external forcing term, and the full Maxwell-Stefan equation for ideal mixtures at low Knudsen numbers. A novel boundary scheme for impermeable solid walls is also suggested to ensure proper mass conservation while effectively managing the spatial interpolations required for multicomponent mixtures with varying molecular masses. We demonstrated the physical consistency and accuracy of the proposed forcing approach through simulations of the ultracentrifuge separation of uranium isotopes and the Loschmidt tube with gravitational effects. Our approach encompasses advanced modeling of species dynamics influenced by force fields, such as those encountered in geological CO_2 sequestration in aquifers and oil reservoirs under gravitational fields.