2026-10-20 | | Total: 20
Discrete adjoint methods give the sensitivities that aerodynamic design and data assimilation require, but they remain hard to obtain for flow solvers that run on graphics processors. This paper presents CuAD, a source-to-source automatic differentiation tool that reads CUDA kernels and writes their tangent and adjoint versions as ordinary CUDA source. Its reverse mode is organised around the GPU thread and exploits the structure of flow-solver kernels: pure accumulations need no tape, the remaining intermediate values are recomputed rather than stored, and adjoint updates are atomic only where several threads reach the same memory. We evaluate CuAD on four kernels from simple flow models: a viscous Burgers residual, a two-dimensional advection-diffusion residual, a quasi-one-dimensional Euler residual and a Biot-Savart summation over point vortices. All generated derivatives agree with finite differences and pass adjoint dot-product tests to round-off, while the adjoint kernels cost two to six times the primal kernels. Compared with the compiler-based tool Clad under the same compiler, CuAD gives identical results at identical cost on the stencil kernels and is 16 to 470 times faster on the summation kernel because it stores no tape. A steady Burgers study shows that in single precision the adjoint gradient of a finite-difference Jacobian reaches errors of 45 to 75 per cent once the condition number grows to $2.5\times10^{5}$, whereas the CuAD Jacobian keeps the error below $5\times10^{-5}$. Four applications complete the evaluation: adjoint fields of the Burgers shock, the sensitivity of a scalar sensor in a cellular vortex, the inverse design of a transonic nozzle whose area is recovered to $2\times10^{-8}$, and an unsteady adjoint of the roll-up of a vortex sheet over 400 time steps.
We derive discrete moment equations exactly equivalent to the lattice Boltzmann method (LBM) and identify the general moment closure shared by kinetic methods based on the discrete-velocity Boltzmann equation. This closure provides a route to correct lower-order dynamics without requiring accurate reconstruction of the higher-order moments, thereby substantially broadening the scope of existing closure strategies that seek increasingly accurate representations of these moments. These advances establish a unified moment framework bridging LBM and moment methods. The framework enables efficient simulations across all Knudsen regimes, overcoming the prohibitive complexity of existing moment methods at high Knudsen numbers. It further provides a common basis for integrating and extending key features of diverse kinetic methods.
This work presents the development and assessment of a hybrid framework coupling numerical simulation, experimental samples, data assimilation (DA) and machine learning (ML) to investigate the flow within an axial compressor stage. The methodology first implements a Multi-Fidelity Ensemble Kalman Filter (MFEnKF) which integrates experimental PIV data with a two-fidelity numerical simulation setup, combining steady-state RANS on a fixed grid with unsteady hybrid RANS/LES simulations on a moving grid. This data assimilation step successfully reduces discrepancies between experiments and CFD realizations. Besides, it provides additional high fidelity samples when compared with the available PIV data. This enriched database is used to train machine learning (ML) tools, with the objective to obtain black-box forcing terms to be included in the discretized form of the Navier-Stokes equations. These models are expected to preserve the accuracy of the DA results while not longer requiring the input of experimental data. The neural network architecture used here maps local physical features such as the velocity, the velocity gradients and the turbulent viscosity, to obtain the volume forcing terms. The analysis of the performance of such ML models shows that global accuracy is preserved but localized increased errors in specific regions of the compressor, such as close to the blade tips, can exceed those observed from the baseline RANS simulations. To resolve this issue, a Mixture of Experts (MoE) method is employed, which classifies and combines the trained ML tools for different flow regions. The discrepancy with experimental data of this method is systematically lower than the one observed for the baseline RANS solution and very similar to the one obtained with the MFEnKF.
Stagnation curves of arbitrary shape (zeros of $\mathbf v(\mathbf x,t)$) are non-generic yet common; we derive their motion. Differentiating $\mathbf v=\mathbf0$ gives $\mathbf L\mathbf U=-\mathbf a$ ($\mathbf U$ curve velocity, $\mathbf a=\mathbf v_t$, $\mathbf L=\nabla\mathbf v$ kills tangent, acts on normal plane via $2\times2$ block $\mathbf K$); dynamics enters only via $\mathbf a$. Normal velocity: $-\mathbf K^{-1}$ times normal $\mathbf a$; tangential solvability: scalar condition $Σ=0$ per point ($Σ$: part of $\mathbf a$ outside range $\mathbf L$). Potential flow: $\mathbf K$ is strain rate, normal velocity pressure gradient over principal strain rate $ε$, persistence needs uniform pressure on it. Stokes flow: tangential vorticity enters $\mathbf K$ (centre lines), $Σ$ involves normal vorticity. Navier-Stokes: $Σ$ set by instantaneous flow via nonlocal pressure, viscosity. If $Σ\ne0$, line becomes, to first order in $t$, isolated points at zeros of $Σ$ (critical points of $\varphi_t$ if potential; maybe none in Stokes flow: loops vanish). Symmetry (line an orbit of rigid-motion group, no velocity along it) suffices for persistence, not necessary: flows steady in rigidly moving frame have $Σ\equiv0$. Incompressibility enters via $\mathrm{tr}\mathbf K=0$; for inviscid irrotational homentropic compressible flow, line dilatation is $\varphi_{tt}/c^2$, so lines can be nodes, persistence needs uniform enthalpy, singular $\mathbf K$ arises at ordinary points, acoustic standing waves give exact examples (node line breaking into alternating sinks, saddles); also degenerate $\mathbf K$ (pairs born at distance $O(\sqrt t)$), curvature, torsion evolution, closure for moving-singularity-driven flows, helical and elliptically conical stagnation-point flow, exact Navier-Stokes example: saddle, centre lines split into four points per period.
The settling of particles in a fluid medium can trigger complex collective dynamics, dramatically affecting the distribution of energy across flow scales. In this work we present the first fully-resolved observations of the turbulence induced by large particle structures, also known as cluster-induced-turbulence (CIT), obtained through particle-resolved direct numerical simulations at high volume fractions and density ratios in large domains. We find that particles arrange in macroscopic fractal structures, exciting turbulent fluctuations at scales much larger than the individual constituents. By virtue of their fractal nature, these clusters force the continuous-phase in a multiscale, self-similar way. We further show that these cluster-driven, vertically decorrelated fluctuations at large scales coexist with, but are fundamentally distinct from, the established particle-driven high-wavenumber scaling. Beyond unveiling a different form of clustering in particle-settling, these results highlight a naturally arising, self-assembling path to fractal-generated turbulence.
We investigate the microstructural dynamics of wormlike micellar (WLM) solutions based on CPyCl/NaSal/D\(_2\)O under shear flow using spatially resolved rheo-diffusion NMR, velocity-compensated diffusion measurements, and rheo-optical imaging in a Taylor--Couette geometry. Two viscoelastic systems were examined: a non-shear-banding WLM with a molar salt-surfactant concentration ratio of \(R=0.4\) and a shear-banding WLM with \(R=0.53\). Rheo-optical and velocity measurements confirm homogeneous flow for the non-shear-banding system and the coexistence of low- and high-shear bands within the stress-plateau regime for the shear-banding system. Despite the pronounced heterogeneity in velocity and shear rate associated with shear banding, the local surfactant self-diffusivity remains spatially uniform across the gap and throughout the stress-plateau regime. Furthermore, analysis of the mean-squared-displacement scaling reveals no measurable difference between the high- and low-shear bands. These results demonstrate that the micellar microstructure remains largely unchanged despite the formation of shear bands and provide direct experimental evidence that flow-induced micellar chain scission is not a dominant mechanism governing shear banding in this WLM system.
Longitudinal dispersion in a yielding fluid depends on the velocity distribution and on how the flow is maintained. We study a neutral tracer in pressure-assisted electrokinetic and magnetohydrodynamic flow of a Casson fluid through a slit microchannel. The nonlinear stress and velocity equations determine the yielded layers, the unyielded core and the axial flow. Mei's multiple-scale homogenization of the advection-diffusion equation then separates transverse relaxation, mean drift and longitudinal spreading, yielding the Taylor-Aris auxiliary problem and effective diffusivity. An exact pressure-driven solution gives cubic decay of the mean velocity and eighth-power decay of the dispersion enhancement near flow arrest. Under fixed pressure and electrical forcing, increasing the yield parameter from $Y=0$ to $0.7$ moves the optimum Hartmann number from $Ha_p=0.696$ to $2.634$, while the corresponding normalized diffusivity decreases from $D_T^{\mathrm C}=1.454$ to $1.015$. At fixed mean throughput and $Ha_p=1$, the trend reverses: $D_T^{\mathrm C}$ increases from $1.000433$ to $1.002311$. The homogenized energy identity traces this difference to the velocity departure from the sectional mean, including that within the unyielded core. Analytical limiting solutions, direct concentration calculations, numerical convergence studies and a coupled Casson shear test support the reduced-flow predictions. An electrolyte-based dimensional assessment identifies the magnetic optima as theoretical forcing states for representative aqueous microchannels. These results are relevant to predicting longitudinal tracer spreading in microchannels carrying fluids with a yield stress, whether pressure or throughput is controlled.
A mechanistic interpretation for baroclinic normal mode instability with exponential growth is the interaction arising from Counter-propagating Rossby Waves (CRWs), where pairs of CRWs phase-lock and interfere constructively. On the other hand, an arbitrary initial condition can lead to finite transient growth as long as there is constructive interference. The present work investigates the characteristics of finite time transient instabilities and its description in terms of CRWs in a baroclinic system with asymmetric interaction, in contrast to usual settings where there is an inherent symmetry that leads to equal and opposite CRWs. A sloped Eady problem on a $f$-plane is considered here, which has the benefit that there are only two CRWs in the system, and the presence of a bottom topographic slope can be formulated in a such a way that only one of the CRW is affected and controlled by a single parameter representing the asymmetry, via modifying the bottom background potential vorticity gradient. We derive conditions where the instantaneous growths have key CRW properties that are independent of the asymmetry parameter for a reasonably general family of norms, and that the instability characteristic in the total energy norm is entirely independent of the asymmetry parameter. For finite time transient growth, the optimal initial condition is in some sense counter to that of the exponential growing normal mode: where the bottom slope is such that it is strengthening the bottom CRW, the normal mode has a larger bottom CRW, but the finite time optimal initial condition wants the bottom CRW to be weak instead, and vice-versa. While the results presented here are for the sloped Eady problem, the same results were also found for an analogous Phillips problem, and we speculate the instability characteristics may have origins in the underlying symmetries that are common to those two problems.
Coupled level-set and volume-of-fluid methods combine the mass conservation of a transported volume fraction with the smooth interface description of a signed-distance function. In the simple coupled variant, the level set is rebuilt from the volume fraction during every time step rather than transported, and it supplies the interface curvature and the fluid properties. This paper presents an implementation of the simple coupled method in OpenFOAM and sets out four conditions that such an implementation must satisfy to stay consistent. The density in the time derivative and in the convective mass flux of the momentum equation must be the same field, the curvature in the surface tension force must come from the level set, the blended kinematic viscosity must reproduce the blended dynamic viscosity, and the smoothed fields must be evaluated on boundary and processor patches as well as in the cells. The solver is tested on the second two-dimensional rising-bubble case of Hysing et al. on two meshes. Up to t= 1.75 s, while the three reference codes of that benchmark agree, the rise velocity differs from the reference by at most 0.0064 and 0.0073 m/s on the two meshes, and its first maximum is 1.3% and 2.0% below the reference. These differences grow slightly with refinement. The same happens when the curvature and the properties are taken from the volume fraction, which reproduces the standard OpenFOAM solver, so their cause lies outside the coupling. Over the same interval the level-set coupling stays closer to the reference than this volume-fraction coupling, with a largest velocity difference of 0.0073 against 0.0094 m/s on the finer mesh. The bubble volume is conserved to the printed precision, and a run on four processors matches the serial run. The solver is released under the GNU General Public License.
Bolus formation is one of the most distinctive features of peristaltic transport, yet the mechanisms governing its development remain incompletely understood. In this work, axisymmetric peristaltic-wave-driven flows under zero-pressure-gradient conditions are investigated using a high-fidelity lattice Boltzmann framework over a broad range of wave amplitudes, wavelengths, and Reynolds numbers spanning the Stokes limit to Re=100. In the Stokes limit, bolus formation is governed by the classical trapping mechanism and exhibits a remarkable wavelength-normalized similarity, with both bolus morphology and transport characteristics approaching common asymptotic behaviors as the wavelength increases. As fluid inertia becomes significant, this similarity progressively breaks down and wavelength-dependent trapping behavior emerges. In the short-wavelength regime, the pronounced geometric bending of the wave reorganizes the local flow structure and produces a distinct off-axis bolus associated with a curvature-induced trapping mechanism. The influence of this mechanism expands toward longer wavelengths with increasing Reynolds number and occupies an increasingly large portion of the parameter space considered in the present study. The results reveal that bolus formation cannot be described by a single universal mechanism. Instead, the observed trapping behavior arises from the competition between classical trapping and curvature-induced trapping, whose relative importance is controlled by wavelength, wave amplitude, and Reynolds number. These findings establish a unified physical framework for understanding bolus formation and transport in peristaltic-wave-driven flows.
Our experiments demonstrate the strong stabilizing effect of a viscoelastic shell on flow-focused viscous jets. The results suggest that stabilization occurs when the polymer viscosity, $η_p$, exceeds a critical threshold, provided the extensional relaxation time, $λ$, is large enough to allow the polymer molecules to undergo the coil-stretch transition. Under these conditions, substantial extensional viscoelastic stresses develop as the liquid accelerates. Since both $η_p$ and $λ$ increase with shell viscosity, the polymer concentration required to stabilize the viscous jet decreases accordingly. Remarkably, moderately to highly viscous jets can be stabilized with very small polymer concentrations, for which no beads-on-a-string (BOAS) instability is observed.
Recent research revealed the orientation-dependent propulsion of a cone-shaped colloidal particle that is exposed to a planar traveling ultrasound wave. Here, we extend the previous research by considering nano- and microcones with different aspect ratios and studying how the propulsion of a particle depends on its orientation and aspect ratio. We also study how the orientation-averaged propulsion of a cone-shaped particle, which corresponds to an isotropic ultrasound field, depends on its aspect ratio and identify an aspect ratio of 1/2 where the orientation-averaged propulsion is particularly strong. To make our simulation results easier reusable for follow-up research, we provide a corresponding simple analytic representation.
The desert-dwelling sandfish (Scincus scincus) moves within dry sand, a material that displays solid and fluidlike behavior. High-speed x-ray imaging shows that below the surface, the lizard no longer uses limbs for propulsion but generates thrust to overcome drag by propagating an undulatory traveling wave down the body. Although viscous hydrodynamics can predict swimming speed in fluids such as water, an equivalent theory for granular drag is not available. To predict sandfish swimming speed, we developed an empirical model by measuring granular drag force on a small cylinder oriented at different angles relative to the displacement direction and summing these forces over the animal movement profile. The agreement between model and experiment implies that the noninertial swimming occurs in a frictional fluid.
A sudden loss of vacuum in a superconducting radio-frequency (SRF) accelerator drives warm gas into the evacuated beamline, where it propagates through interconnected SRF cavities cooled by liquid helium, causing surface contamination, intense heat deposition, and dangerous pressure buildup. Yet predicting how such a failure propagates through realistic multicell cavities has remained difficult because compressible gas dynamics, condensation and cryogenic heat transfer are strongly coupled. Here we develop and experimentally validate a multidimensional computational framework that resolves these processes, and apply it to a full-scale nine-cell SRF cavity. We find that the repeated cell geometry produces staircase-like gas-front propagation, as impingement, recirculation and localized cryopumping strongly retard the bulk flow within each cell. Across a broad range of breach sizes, the cavity traversal time follows a robust power-law scaling with leakage area, providing quantitative guidance for the response time of beamline isolation and protection systems. The simulations also reveal a geometry-confined Mach disk whose position and size track the advancing gas front, suggesting that its acoustic emission could potentially enable rapid, non-invasive detection and localization of vacuum failures. These findings pave the way for a new generation of predictive diagnostics and protection strategies to safeguard large-scale superconducting accelerators.
We present a simple and robust method for generating freely suspended liquid sheets by directing a single laminar liquid jet onto the sharp edge of an inclined plate. By optimizing liquid composition, impact geometry, and flow conditions, we identify a 60:40 ethylene-glycol-deionized-water volumetric mixture for stable sheet formation. The resulting sheets span 8 mm in height and 3 mm in width, with thicknesses from 1.1 to 4.8 micrometers measured by spatial and spectral interferometry. Operated in vacuum as continuously renewed plasma mirrors, these targets demonstrate stable performance from 0.5 to 10 kHz using 175-fs Yb:KGW laser pulses, reaching a maximum S-polarized reflectivity of 66.5% at an intensity of 4 x 10^14 W/cm^2. At higher intensities using 25-fs Ti:sapphire pulses at 1 kHz, maximum S-polarized reflectivity reaches 77% at 2 x 10^16 W/cm^2. The reflected beam exhibits high temporal stability, with angular pointing fluctuations of 18.19 +/- 0.53 microradians. Our results demonstrate a tunable, continuously renewed liquid-sheet target for multi-kHz laser-plasma experiments.
Nearby relaxation rates make two membrane viscosities difficult to distinguish from a shape transient. We show how calibrated preparation removes this loss of local sensitivity in a linear Kelvin-Voigt sphere with finite area elasticity and intrinsic membrane dissipation. Two independent releases, with a known signed amplitude ratio and calibrated relative gain and timing, yield normal-displacement records whose fixed linear combination separates the shear and dilatational rates. Their common amplitude remains unknown. For fixed sampling, positive-definite weights, bounded positive rates and nonzero excitation, the amplitude-projected sensitivity has a positive lower bound even at rate coincidence. Without relative calibration, the paired sensitivity decreases linearly with the rate gap; the single tangential-release record decreases quadratically. The calibrated bound is compatible with fixed finite initial energy. A positive local bound also persists with independent ratio calibration of fixed positive precision under a Gaussian error model. Supplied elastic coefficients convert the recovered rates into the two surface viscosities. An unbounded spherical Stokes extension quantifies how bulk dissipation changes the modal patterns and provides a sufficient full-Jacobian criterion for retaining local material sensitivity.
Residual stress remains in an object even in the absence of external forces or thermal pressure, which, in turn, may cause significant plastic deformations. In case the residual stress creates unwanted effects on the material and so is undesirable, an efficient solution is necessary to track and eliminate this stress. Smoothed Particle Hydrodynamics has been extensively used in solid mechanics simulations and the inherent colour-field generation approach is a promising tracker for the residual stress. In this paper, we propose a way to use the colour-field approach for eliminating the residual stress and prevent the undesirable premature failure of solid objects.
Integrating condensed noble gases with superconducting circuits offers new opportunities for operating hybrid quantum devices. In particular, electrons on solid neon have recently emerged as exceptionally coherent charge qubits, however the growth of the underlying neon films on patterned devices remains poorly controlled. We monitor neon films near the triple point using a multimode resonator fabricated from high-$T_c$ YBCO, which remains superconducting at temperatures where the liquid phase exists and whose geometry sets a thickness-sensing window near 1 $μ$m. In the liquid state we observe rich film dynamics: uniform wetting that saturates near 20 nm, drainage of thicker films, and spontaneous thickness oscillations, reflecting the extreme sensitivity of thick films in this regime. A galvanic connection to our sensing volume enables in-situ electrostatic tuning of the liquid film thickness. Statistics of freezing the neon across various liquid levels reveal variations of the solid film thickness by approximately an order of magnitude. We also observe a two-fold increase in the quality factor of the resonators at the level of $10^3$ with added neon and rule out dielectric loss as the limiting factor. Our results demonstrate that hermetic packaging facilitates the investigation of both rich non-equilibrium fluid dynamics and the impact of condensed noble gas films on microwave loss in superconducting circuits, but may not be the best modality for growing solid neon films with precisely controlled thickness.
Low-order Carleman embeddings replace nonlinear dynamics by a larger linear system, but truncation errors can accumulate even when the underlying nonlinear model is accurately approximated by a quadratic equation. We construct a residual Carleman lift (RC2) of lattice Boltzmann dynamics in which an independently evolved spectral background supplies the coefficients and forcing of a second-order residual lift. In a double-shear-layer test, conventional fixed-rest Carleman reaches $87.3\%$ final error, whereas prepared RC2 achieves $0.136\%$ through vortex roll-up, without restarting the lift or using fine-reference feedback. The results demonstrate control of the finite-time accuracy of a fixed-order linear embedding through nonlinear preparation.
We propose using a directional time-varying (rotating) driving magnetic field to suppress magneto-Rayleigh-Taylor (MRT) instability in dynamic Z-pinches. A rotational drive magnetic field is equivalent to two magnetic-field components, {\Theta} and Z, that alternate in time, referred to as an alternate Theta-Z-pinch configuration. We consider the finitely thick cylindrical liner configuration in this paper. We numerically integrate the perturbation equation to stagnation time based on the optimal background unperturbed trajectories. We assess the cumulative growth of the dominant mode selected by some mechanism at the beginning of an implosion. The maximum e-folding number at stagnation of the dominant mode of an optimized alternate Theta-Z-pinch is significantly lower than that of the standard Theta- or Z-pinch. The directional rotation of the magnetic field contributes to suppress the instabilities, independent of the finite thickness. The finite thickness effect plays a role only when the orientation of the magnetic field varies in time whereas it does not appear in the standard Theta- or Z-pinch. The rotating frequency of the magnetic field and the thickness of liner are both having a monotonic effect on suppression. Their synergistic effect can enhance the suppression on MRT instability. Because the MRT instability can be well suppressed in this way, the alternate Theta-Z-pinch configuration has potential applications in liner inertial fusion. This work is supported by the NSFC (Grant Nos. 11405167, 51407171, 11571293, 11605188, and 11605189) and the Foundation of the China Academy of Engineering Physics (No. 2015B0201023).