In deep learning, interval neural networks are used to quantify the uncertainty of a pre-trained neural network. Suppose we are given a computational problem $P$ and a pre-trained neural network $\Phi _P$ that aims to solve $P$. An interval neural network is then a pair of neural networks $(\underline {\phi }, \overline {\phi })$, with the property that $\underline {\phi }(y) \leq \Phi _P(y) \leq \overline {\phi }(y)$ for all inputs $y$, where the inequalities are meant componentwise. $(\underline {\phi }, \overline {\phi })$ are specifically trained to quantify the uncertainty of $\Phi _P$, in the sense that the size of the interval $[\underline {\phi }(y),\overline {\phi }(y)]$ quantifies the uncertainty of the prediction $\Phi _P(y)$. In this paper, we investigate the phenomenon when algorithms cannot compute interval neural networks in the setting of inverse problems. We show that in the typical setting of a linear inverse problem, the problem of constructing an optimal pair of interval neural networks is non-computable, even with the assumption that the pre-trained neural network $\Phi _P$ is an optimal solution. In other words, there exist classes of training sets $\Omega$, such that there is no algorithm, even randomised (with probability $p \geq 1/2$), that computes an optimal pair of interval neural networks for each training set ${\mathcal{T}} \in \Omega$. This phenomenon happens even when we are given a pre-trained neural network $\Phi _{{\mathcal{T}}}$ that is optimal for $\mathcal{T}$. This phenomenon is intimately linked to instability in deep learning.

We study the stability of a steady Eckart streaming jet flowing in a closed cylindrical cavity. This configuration is a generic representation of industrial processes where driving flows in a cavity by means of acoustic forcing offers a contactless way of stirring or controlling flows. Successfully doing so, however, requires sufficient insight into the topology induced by the acoustic beam. This, in turn, raises the more fundamental question of whether the basic jet topology is stable and, when it is not, of the alternative states that end up being acoustically forced. To answer these questions, we consider a flow forced by an axisymmetric diffracting beam of attenuated sound waves emitted by a plane circular transducer at one cavity end. At the opposite end, the jet impingement drives recirculating structures spanning nearly the entire cavity radius. We rely on linear stability analysis (LSA) together with three-dimensional nonlinear simulations to identify the flow destabilisation mechanisms and to determine the bifurcation criticalities. We show that flow destabilisation is closely related to the impingement-driven recirculating structures, and that the ratio $C_R$ between the cavity and the maximum beam radii plays a key role on the flow stability. In total, we identified four mode types destabilising the flow. For $4 \leqslant C_R \leqslant 6$, a non-oscillatory perturbation rooted in the jet impingement triggers a supercritical bifurcation. For $C_R = 3$, the flow destabilises through a subcritical non-oscillatory bifurcation and we explain the topological change of the unstable perturbation by analysing its critical points. Further reducing $C_R$ increases the shear within the flow and gradually moves the instability origin to the shear layer between impingement-induced vortices: for $C_R = 2$, an unstable travelling wave grows out of a subcritical bifurcation, which becomes supercritical for $C_R=1$. For each geometry, the nonlinear three-dimensional (3-D) simulations confirm both the topology and the growth rate of the unstable perturbation returned by LSA. This study offers fundamental insight into the stability of acoustically driven flows in general, but also opens possible pathways to either induce turbulence acoustically or to avoid it in realistic configurations.

The effects of Reynolds number across ${\textit{Re}}=1000$, $2500$, $5000$ and $10\,000$ on separated flow over a two-dimensional NACA0012 airfoil at an angle of attack of $\alpha =14^\circ$ are investigated through biglobal resolvent analysis. We identify modal structures and energy amplifications over a range of frequencies, spanwise wavenumbers, and values of the discount parameter, providing insights across various time scales. Using temporal discounting, we find that the shear-layer dynamics dominates over short time horizons, while the wake dynamics becomes the primary amplification mechanism over long time horizons. Spanwise effects also appear over long time horizons, sustained by low frequencies. The low-frequency and high-wavenumber structures are found to be dominated by elliptic mechanisms within the recirculation region. At a fixed angle of attack and across the Reynolds numbers, the response modes shift from wake-dominated structures at low frequencies to shear-layer-dominated structures at higher frequencies. The frequency at which the dominant mechanism changes is independent of the Reynolds number. Comparisons at a different angle of attack ($\alpha =9^\circ$) show that the transition from wake to shear-layer dynamics with increasing frequency only occurs if the unsteady flow is three-dimensional. We also study the dominant frequencies associated with wake and shear-layer dynamics across the angles of attack and Reynolds numbers, and confirm characteristic scaling laws from the literature.

Direct numerical simulations of a uniform flow past a fixed spherical droplet are performed to determine the parameter range within which the axisymmetric flow becomes unstable. The problem is governed by three dimensionless parameters: the drop-to-fluid dynamic viscosity ratio, $\mu ^\ast$, and the external and internal Reynolds numbers, ${\textit{Re}}^e$ and ${\textit{Re}}^i$, which are defined using the kinematic viscosities of the external and internal fluids, respectively. The present study confirms the existence of a regime at low-to-moderate viscosity ratio where the axisymmetric flow breaks down due to an internal flow instability. In the initial stages of this bifurcation, the external flow remains axisymmetric, while the asymmetry is generated and grows only inside the droplet. As the disturbance propagates outward, the entire flow first transits to a biplanar-symmetric flow, characterised by two pairs of counter-rotating streamwise vortices in the wake. A detailed examination of the flow field reveals that the vorticity on the internal side of the droplet interface is driving the flow instability. Specifically, the bifurcation sets in once the maximum internal vorticity exceeds a critical value that decreases with increasing ${\textit{Re}}^i$. For sufficiently large ${\textit{Re}}^i$, internal flow bifurcation may occur at viscosity ratios of $\mu ^\ast = {\mathcal{O}}(10)$, an order of magnitude higher than previously reported values. Finally, we demonstrate that the internal flow bifurcation in the configuration of a fixed droplet in a uniform fluid stream is closely related to the first path instability experienced by a buoyant, deformable droplet of low-to-moderate $\mu ^\ast$ freely rising in a stagnant liquid.

A literature review suggests that the flows past simply connected bodies with aspect ratio close to unity and symmetries aligned with the flow follow a consistent sequence of regimes (steady, periodic, quasiperiodic) as the Reynolds number increases. However, evidence is fragmented, and studies are rarely conducted using comparable numerical or experimental set-ups. This paper investigates the wake dynamics of two canonical bluff bodies with distinct symmetries: a cube (discrete) and a sphere (continuous). Employing three-dimensional (3-D) global linear stability analysis and nonlinear simulations within a unified numerical framework, we identify the bifurcation sequence driving these regime transitions. The sequence: a pitchfork bifurcation breaks spatial symmetry; a Hopf bifurcation introduces temporal periodicity ($St_1$); a Neimark–Sacker bifurcation destabilises the periodic orbit, leading to quasiperiodic dynamics with two incommensurate frequencies ($St_1, St_2$). A Newton–Krylov method computes the unstable steady and periodic base flows without imposing symmetry constraints. Linear stability reveals similarities between the cube and sphere in the spatial structure of the leading eigenvectors and in the eigenvalue trajectories approaching instability. This study provides the first confirmation of a Neimark–Sacker bifurcation to quasiperiodicity in these 3-D wakes, using Floquet stability analysis of computed unstable periodic orbits and their Floquet modes. The quasiperiodic regime is described in space and time by the Floquet modes’ effects on the base flow and a spectrum dominated by the two incommensurate frequencies and tones arising from nonlinear interactions. Although demonstrated for a cube and a sphere, this bifurcation sequence, leading from steady state to quasiperiodic dynamics, suggests broader applicability beyond these geometries.

The primary bifurcation of the flow past three-dimensional axisymmetric bodies is investigated. We show that the azimuthal vorticity generated at the body surface is at the root of the instability, and that the mechanism proposed by Magnaudet & Mougin (2007, J. Fluid Mech., vol. 572, 311–337) in the context of spheroidal bubbles extends to axisymmetric bodies with a no-slip surface. The instability arises in a thin region of the flow in the near wake, and is associated with the occurrence of strong vorticity gradients. We propose a simple yet effective scaling law for the prediction of the instability, based on a measure of the near-wake vorticity and of the radial extent of the separation bubble. At criticality, the resulting Reynolds number collapses approximately to a constant value for bodies with different geometries and aspect ratios, with a relative variation that is one order of magnitude smaller than that of the standard Reynolds number based on the free-stream velocity and body diameter. The new scaling can be useful to assess whether the steady flow past axisymmetric bodies is globally unstable, without the need for an additional stability analysis.

Yaw control can effectively enhance wind farm power output, but the vorticity distribution and coherent structures in yawed turbine wakes remain poorly understood. We propose a physical model capable of accurately predicting tip vortex dynamics from their generation to destabilisation. This model integrates a point vortex framework with advanced blade element momentum theory and vortex cylinder theory for yawed turbines. Comparisons with large eddy simulations demonstrate that the model effectively predicts the vorticity distribution of tip vortices and the wake profile of yawed turbines. Finally, we employ sparsity-promoting dynamic mode decomposition to analyse the dynamics of the far wake. Our analysis reveals four primary mode types: (i) the averaged mode; (ii) shear modes; (iii) harmonic modes; and (iv) merging modes. Under yawed conditions, these modes become asymmetric, leading to interactions between the tip and root vortex modes. This direct interaction plays a critical role during the formation process of the counter-rotating vortex pair observed in yawed wakes.

This study investigated the cylindrically divergent Rayleigh–Taylor instability (RTI) on a liquid–gas interface and its dependence on initial conditions. A novel hydrophobic technique was developed to generate a two-dimensional water–air interface with controlled initial conditions. The experimental configuration utilised high-pressure air injection to produce uniform circumferential acceleration. Amplitude measurements over time revealed that the cylindrical RTI growth depends strongly on the azimuthal wavenumber. Experimental results demonstrated that surface tension significantly suppresses the liquid–gas cylindrical RTI, even inducing a freeze-out and oscillatory perturbation growth – a phenomenon observed for the first time. Spectrum analysis of the interface contours demonstrated that the cylindrical RTI evolves in a weakly nonlinear regime. Linear and weakly nonlinear models were derived to accurately predict the time-varying interface amplitudes and high-order modes. The linear model was further used to determine conditions for unstable, freeze-out and oscillatory solutions of the cylindrically divergent RTI. These findings offer valuable insights into manipulating hydrodynamic instabilities in contracting/expanding geometries using surface tension.

The linear stability of a thermally stratified fluid layer between horizontal walls, where non-Brownian thermal particles are injected continuously at one boundary and extracted at the other – a system known as particulate Rayleigh–Bénard (pRB) – is studied. For a fixed volumetric particle flux and minimal thermal coupling, reducing the injection velocity stabilises the system when heavy particles are introduced from above, but destabilises it when light particles are injected from below. For very light particles (bubbles), low injection velocities can shift the onset of convection to negative Rayleigh numbers, i.e. heating from above. Particles accumulate non-uniformly near the extraction wall and in regions of strong vertical flow, aligning with either wall-impinging or wall-detaching zones depending on whether injection is at sub- or super-terminal velocity. The increase of the volumetric particle flux always enhances these effects.

Phase change materials (PCMs) hold considerable promise for thermal energy storage applications. However, designing a PCM system to meet a specific performance presents a formidable challenge, given the intricate influence of multiple factors on the performance. To address this challenge, we hereby develop a theoretical framework that elucidates the melting process of PCMs. By integrating stability analysis with theoretical modelling, we derive a transition criterion to demarcate different melting regimes, and subsequently formulate the melting curve that uniquely characterises the performance of an exemplary PCM system. This theoretical melting curve captures the key trends observed in experimental and numerical data across a broad parameter space, establishing a convenient and quantitative relationship between design parameters and system performance. Furthermore, we demonstrate the versatility of the theoretical framework across diverse configurations. Overall, our findings deepen the understanding of thermo-hydrodynamics in melting PCMs, thereby facilitating the evaluation, design and enhancement of PCM systems.

Experimental investigation of the Rayleigh–Taylor instability (RTI) and its dependence on initial conditions has been challenging, primarily due to the difficulty of creating a well-defined gaseous interface. To address this, a novel soap film technique was developed to create a discontinuous two-dimensional SF$_6$air interface with precisely controlled initial conditions. High-order modes were superimposed on a long-wavelength perturbation to study the influence of initial conditions on RTI evolution. Experiments conducted at Atwood numbers ranging from 0.26 to 0.66 revealed that bubble growth shows a weak dependence on both initial conditions and Atwood numbers, whereas spike growth is more influenced by these factors. Spike growth accelerates as the wavenumber of the imposed high-order modes decreases and/or the Atwood number increases. To quantify these effects, a variation on the previously developed potential flow model was applied, capturing the suppression of high-order modes and Atwood number dependence on RTI growth. In turbulent flow, the self-similar factors of bubbles and spikes exhibit minimal sensitivity to initial conditions. However, in relation to the Atwood number, the self-similar factors of bubbles (or spikes) demonstrate negligible (or significant) dependence. Comparisons with literature revealed that two-dimensional flows yield lower self-similar factors than three-dimensional flows. Furthermore, the discontinuity of the initial interface in this study, achieved through the soap film technique, results in faster spike growth compared with previous studies involving a diffusive initial interface. These findings provide critical insights into the nonlinear dynamics of RTI and underscore the importance of well-characterised initial conditions in experimental studies.

This paper explores the construction of quadratic Lyapunov functions for establishing the conditional stability of shear flows described by truncated ordinary differential equations, addressing the limitations of traditional methods like the Reynolds–Orr equation and linear stability analysis. The Reynolds–Orr equation, while effective for predicting unconditional stability thresholds in shear flows due to the non-contribution of nonlinear terms, often underestimates critical Reynolds numbers. Linear stability analysis, conversely, can yield impractically high limits due to subcritical transitions. Quadratic Lyapunov functions offer a promising alternative, capable of proving conditional stability, albeit with challenges in their construction. Typically, sum-of-squares programs are employed for this purpose, but these can result in sizable optimisation problems as system complexity increases. This study introduces a novel approach using linear transformations described by matrices to define quadratic Lyapunov functions, validated through nonlinear optimisation techniques. This method proves particularly advantageous for large systems by leveraging analytical gradients in the optimisation process. Two construction methods are proposed: one based on general optimisation of transformation matrix coefficients, and another focusing solely on the system’s linear aspects for more efficient Lyapunov function construction. These approaches are tested on low-order models of subcritical transition and a two-dimensional Poiseuille flow model with degrees of freedom nearing 1000, demonstrating their effectiveness and efficiency compared with sum-of-squares programs.

The evolution of a Lamb–Oseen vortex is studied in a stratified rotating fluid under the complete Coriolis force. In a companion paper, it was shown that the non-traditional Coriolis force generates a vertical velocity field and a vertical vorticity anomaly at a critical radius when the Froude number is larger than unity. Below a critical non-traditional Rossby number $\widetilde {Ro}$, a two-dimensional shear instability was next triggered by the vorticity anomaly. Here, we test the robustness of this two-dimensional evolution against small three-dimensional perturbations. Direct numerical simulations (DNS) show that the two-dimensional shear instability then develops only in an intermediate range of non-traditional Rossby numbers for a fixed Reynolds number $Re$. For lower $\widetilde {Ro}$, the instability is three-dimensional. Stability analyses of the flows in the DNS prior to the instability onset fully confirm the existence of these two competing instabilities. In addition, stability analyses of the local theoretical flows at leading order in the critical layer demonstrate that the three-dimensional instability is due to the shear of the vertical velocity. For a given Froude number, its growth rate scales as $Re^{2/3}/\widetilde {Ro}$, whereas the growth rate of the two-dimensional instability depends on $Re/\widetilde {Ro}^2$, provided that the critical layer is smoothed by viscous effects. However, the growth rate of the three-dimensional instability obtained from such local stability analyses agrees quantitatively with those of the DNS flows only if second-order effects due to the traditional Coriolis force and the buoyancy force are taken into account. These effects tend to damp the three-dimensional instability.

Stall cells are transverse cellular patterns that often appear on the suction side of airfoils near stalling conditions. Wind-tunnel experiments on a NACA4412 airfoil at Reynolds number ${Re}=3.5 \times 10^5$ show that they appear for angles of attack larger than $\alpha = 11.5^{\circ }\ (\pm 0.5^{\circ })$. Their onset is further investigated based on global stability analyses of turbulent mean flows computed with the Reynolds-averaged Navier–Stokes (RANS) equations. Using the classical Spalart–Allmaras turbulence model and following Plante et al. (J. Fluid Mech., vol. 908, 2021, A16), we first show that a three-dimensional stationary mode becomes unstable for a critical angle of attack $\alpha = 15.5^{\circ }$ which is much larger than in the experiments. A data-consistent RANS model is then proposed to reinvestigate the onset of these stall cells. Through an adjoint-based data-assimilation approach, several corrections in the turbulence model equation are identified to minimize the differences between assimilated and reference mean-velocity fields, the latter reference field being extracted from direct numerical simulations. Linear stability analysis around the assimilated mean flow obtained with the best correction is performed first using a perturbed eddy-viscosity approach which requires the linearization of both RANS and turbulence model equations. The three-dimensional stationary mode becomes unstable for angle $\alpha = 11^{\circ }$ which is in significantly better agreement with the experimental results. The interest of this perturbed eddy-viscosity approach is demonstrated by comparing with results of two frozen eddy-viscosity approaches that neglect the perturbation of the eddy viscosity. Both approaches predict the primary destabilization of a higher-wavenumber mode which is not experimentally observed. Uncertainties in the stability results are quantified through a sensitivity analysis of the stall cell mode's eigenvalue with respect to residual mean-flow velocity errors. The impact of the correction field on the results of stability analysis is finally assessed.

Understanding interfacial instability in a coflow system has relevance in the effective manipulation of small objects in microfluidic applications. We experimentally elucidate interfacial instability in stratified coflow systems of Newtonian and viscoelastic fluid streams in microfluidic confinements. By performing a linear stability analysis, we derive equations that describe the complex wave speed and the dispersion relationship between wavenumber and angular frequency, thus categorizing the behaviour of the systems into two main regimes: stable (with a flat interface) and unstable (with either a wavy interface or droplet formation). We characterize the regimes in terms of the capillary numbers of the phases in a comprehensive regime plot. We decipher the dependence of interfacial instability on fluidic parameters by decoupling the physics into viscous and elastic components. Remarkably, our findings reveal that elastic stratification can both stabilize and destabilize the flow, depending on the fluid and flow parameters. We also examine droplet formation, which is important for microfluidic applications. Our findings suggest that adjusting the viscous and elastic properties of the fluids can control the transition between wavy and droplet-forming unstable regimes. Our investigation uncovers the physics behind the instability involved in interfacial flows of Newtonian and viscoelastic fluids in general, and the unexplored behaviour of interfacial waves in stratified liquid systems. The present study can lead to a better understanding of the manipulation of small objects and production of droplets in microfluidic coflow systems.

We study the existence and strong instability of standing wave solutions for the fractional nonlinear Schrödinger equation

\begin{equation*}\begin{cases}\mathrm{i} \psi_{t}=\left(-\Delta\right)^s \psi-\left(|\psi|^{p-2}\psi+\mu|\psi|^{q-2}\psi\right), & \quad(t,x)\in(0,\infty)\times\mathbb{R}^N,\\\psi(0,x)=\psi_0(x),&\quad x\in\mathbb{R}^N,\end{cases}\end{equation*}

where $N\geq2$, $0 \lt s \lt 1$, $2 \lt q \lt p \lt 2_s^*=2N/(N-2s)$, and $\mu\in\mathbb{R}$. The primary challenge lies in the inhomogeneity of the nonlinearity.We deal with the following three cases: (i) for $2 \lt q \lt p \lt 2+4s/N$ and µ < 0, there exists a threshold mass a0 for the existence of the least energy normalized solution; (ii) for $2+4s/N \lt q \lt p \lt 2_s^*$ and µ > 0, we reveal the existence of the ground state solution, explore the strong instability of standing waves, and provide a blow-up criterion; (iii) for $2 \lt q\leq2+4s/N \lt p \lt 2_s^*$ and µ < 0, the strong instability of standing wave solutions is demonstrated. These findings are illuminated through variational characterizations, the profile decomposition, and the virial estimate.

The resolvent analysis reveals the worst-case disturbances and the most amplified response in a fluid flow that can develop around a stationary base state. The recent work by Padovan et al. (J. Fluid Mech., vol. 900, 2020, A14) extended the classical resolvent analysis to the harmonic resolvent analysis framework by incorporating the time-varying nature of the base flow. The harmonic resolvent analysis can capture the triadic interactions between perturbations at two different frequencies through a base flow at a particular frequency. The singular values of the harmonic resolvent operator act as a gain between the spatiotemporal forcing and the response provided by the singular vectors. In the current study, we formulate the harmonic resolvent analysis framework for compressible flows based on the linearized Navier–Stokes equation (i.e. operator-based formulation). We validate our approach by applying the technique to the low-Mach-number flow past an airfoil. We further illustrate the application of this method to compressible cavity flows at Mach numbers of 0.6 and 0.8 with a length-to-depth ratio of $2$. For the cavity flow at a Mach number of 0.6, the harmonic resolvent analysis reveals that the nonlinear cross-frequency interactions dominate the amplification of perturbations at frequencies that are harmonics of the leading Rossiter mode in the nonlinear flow. The findings demonstrate a physically consistent representation of an energy transfer from slow-evolving modes toward fast-evolving modes in the flow through cross-frequency interactions. For the cavity flow at a Mach number of 0.8, the analysis also sheds light on the nature of cross-frequency interaction in a cavity flow with two coexisting resonances.