betaFeedback

WorksPhysics and Astronomy

Zero-Mean Oscillations Drive Recirculations in Potential Flows at Vanishing Reynolds Number

Abstract

Recirculating flows are central to transport phenomena, but their generation in potential flows requires additional physical mechanisms. We show experimentally that recirculation at vanishing Reynolds number can be produced at prescribed locations by coupling oscillations of a Hele-Shaw cell ceiling with an off-center passive obstacle. The net drift is well described by a potential flow model with no adjustable parameters. The resulting flows are localized around the obstacles and can be combined to generate programmable flow patterns at the microscale.

Recirculations are ubiquitous across scales and disciplines and play a critical role in material transport and trapping. Examples range from eddy currents in electric conductors [1] to vortices generated by jellyfish to capture prey [2] to planetary-scale vortices such as Jupiter’s Great Red Spot [3]. Recirculations at large Reynolds numbers typically arise from inertia. At small Reynolds numbers, they emerge from the coupling of kinematic boundary conditions with the system geometry through viscous diffusion [4, 5].

In potential flow problems, which describe the motion of viscous fluid in porous media or the depth-averaged flow in the Hele-Shaw geometry, it is classically considered that recirculation is not possible [6]. In these flows, the velocity is proportional to the gradient of the potential. Therefore, the circulation must vanish around any closed contour, even in the presence of an obstacle. Although several studies have recently demonstrated the existence of recirculation in Hele-Shaw cells, they rely on additional physical mechanisms beyond those of potential flows [7, 8, 9, 10]. It remains unclear whether a closed recirculation can emerge from potential flows alone.

Here, we demonstrate with theory and experiments that a recirculation is possible by changing from an Eulerian to a Lagrangian framework. We consider a periodically deformed Hele-Shaw cell containing a fixed obstacle. The resulting oscillatory flow has zero mean and remains potential at all times. Yet, the obstacle rectifies it into a closed and localized Lagrangian recirculation that recalls ratchets in fluidic systems [11, 12] and beyond [13]. We show that the resulting drift is purely kinematic, akin to Stokes drift in inertial traveling waves [14, 15, 16], and that its strength is set by the signed area enclosed by the actuation cycle.

The experimental setup consists of a PDMS (Sylgard 184) microfluidic chip comprising a channel of height h0=500 μmh_0 = 500\ \mathrm{\mu m}, much smaller than its width Ly=3.8 mmL_y = 3.8\ \mathrm{mm} and length Lx=20 mmL_x = 20\ \mathrm{mm}. The channel is flanked by two independent air chambers, of height 2.4 mm and width 7.6 mm, as shown in the schematic in Figure 1(a). Applying negative pressures to the air chambers via a multichannel pressure controller (OB1 MK4, Elveflow) results in the deformation of the entire device. In particular, the ceiling of the microfluidic channel deflects downward, with a deflection amplitude proportional to the applied pressure [17, 18].

Figure 1. (a) Experimental setup: cross-section and top view of the microfluidic chip. (b) Channel ceiling deformation during one pressure actuation cycle, measured by fluorescence height profilometry in the absence of obstacle, for φ=−π2\varphi=-\frac{\pi}{2}. Scale bars: 2 mm. (c) Corresponding transverse height profiles at the center of the microfluidic channel. (d) Stroboscopic image of the flow in the microfluidic channel. Scale bar: 500 μ\mum.

Imposing different negative pressures in each air chamber yields an asymmetric channel ceiling deformation. We concentrate on periodic actuations. A pressure P1(t)=−Po2(1+sin⁡(ωt))P_1(t) = -\frac{P_o}{2}(1 + \sin(\omega t)) is imposed in one air chamber, and a pressure P2(t)=−Po2(1+sin⁡(ωt+φ))P_2(t) = -\frac{P_o}{2}(1 + \sin(\omega t + \varphi)) in the other. Here, PoP_o and ω\omega respectively denote the amplitude and angular frequency of the actuation, and φ\varphi the phase delay between the two actuations. The time evolution of the deformation of the entire microfluidic channel for φ=−π2\varphi= -\frac{\pi}{2} and Po=500P_o = 500 mbar is shown in Figure 1(b). The deformation profile is symmetric along yy when both actuating pressures are equal, and tilted otherwise, with a maximum deformation ≈250 μm\approx250\ \mathrm{\mu m}. Corresponding transverse channel height profiles along yy are shown in Figure 1(c). The tilt direction of the ceiling switches between the two halves of the pressure cycle, depending on the relative magnitudes of P1P_1 and P2P_2.

The microfluidic channel includes five small traps etched in the ceiling of the device and centered along yy [19]. These traps appear as darker dots in Figure 1(b), and as a kink at y=0y = 0 in the profiles shown in Figure 1(c). A fixed obstacle, here a cylindrical hydrogel of radius RR, is locally photopolymerized using a digital micromirror device (Primo, Alvéole) and centered at one of the traps. The hydrogel is composed of a mixture of 55% polyethylene glycol (PEG, MW 200), 15% polyethylene glycol diacrylate (PEGDA, MW 700), and 30% water. Once polymerized, the obstacle adheres to the glass slide and remains at a fixed position even in the presence of actuation. The channel is then filled with PEG, a Newtonian fluid of kinematic viscosity ν=57 mm2/s\nu= 57\ \mathrm{mm^2/s} [20], and seeded with fluorescent particles of diameter 3 μm\mathrm{\mu m} for flow visualization or diameter 500 nm for particle image velocimetry (PIV). The actuation of the air chambers drives the fluid in and out of the channel through the left and right outlets. The resulting flow wraps around the obstacle and changes orientation and direction with the oscillating forcing. Imaging the channel once per actuation cycle reveals that the forcing generates a recirculating net flow localized around the obstacle, as shown in Figure 1(d) for φ=−π2\varphi=-\frac{\pi}{2} (see also Movie S1 in [21]). This stroboscopic image is effectively a Poincaré map: it does not show the full tracer trajectories, but only captures the net flow over multiple cycles. Note that a drift occurs even in the absence of the obstacle due to the spatial variations of the ceiling height. Here, we concentrate on the localized recirculation that results from the presence of the obstacle.

To understand the origin of the net recirculation, the flow profile around the obstacle is reconstructed at each instant using PIV [22]. The experimental streamlines are shown at five instants in time in the top row of Figure 2. Near the obstacle, the flow resembles a uniform flow past a cylinder whose direction rotates due to the phase delay between the actuation sources. The rotation of the streamlines around the obstacle points to the mechanism responsible for the net drift: the flow is zero-mean and time-periodic, but passive tracers experience varying velocities along their trajectories, which leads to an overall displacement per actuation period. This can be seen in Figure 2, where virtual tracer trajectories show a small but systematic offset between their initial (circles) and final (crosses) positions after one period.

Figure 2. Experimental and predicted streamlines around the obstacle. Top row: streamlines deduced from instantaneous PIV measurements between two successive frames at five time instants over one period. Bottom row: predicted streamlines from the potential flow model at the same time instants. Red and blue arrows respectively indicate positive and negative horizontal velocity. The trajectories of four virtual tracers are indicated on all panels (initial positions are indicated with open circles, and current positions with crosses). The rightmost panels show the trajectories over 10 periods.

The net recirculation can be modeled using potential flow theory to describe the flow in the chip, treated as a Hele-Shaw cell [8, 23]. Inertial effects are indeed negligible in these experiments, as both the Reynolds number Re⁡=Uih0ν≈10−3\operatorname{Re}=\frac{U_i h_0}{\nu}\approx10^{-3} (where UiU_i is a characteristic instantaneous velocity) and the Womersley number Wo⁡2=ωh02ν≈10−2\operatorname{Wo}^2=\frac{\omega h_0^2}{\nu}\approx10^{-2} are very small. We first define a channel without obstacle, of dimensions LxL_x and LyL_y. The Eulerian flow field satisfies uE=∇ϕ0\mathbf{u}_{\mathrm{E}}=\nabla\phi_0, where the velocity potential ϕ0\phi_0 obeys the Poisson equation whose source term represents the effects of vertical ceiling actuation throughout the domain:

Δϕ0=1h0∂δ∂t.(1)\Delta\phi_0=\frac{1}{h_0}\frac{\partial\delta}{\partial t}. \tag*{(1)}

Here, δ\delta is the ceiling displacement. Experimentally, the ceiling deformation is not constant along xx, but shows a maximum in the center of the device (x=0x=0) and a minimum at the device edges (x=±Lx/2x=\pm L_x/2), which is modeled as a cosine profile cos⁡(πx/Lx)\cos(\pi x/L_x) (Figure 1(b)). Further, we approximate δ\delta as a linear function of yy (Figure 1(c)). δ\delta is therefore given by:

δ(x,y,t)=[δ1+δ22+δ1−δ2Lyy]cos⁡(kx),(2)\delta(x,y,t)=\left[\frac{\delta_1+\delta_2}{2}+\frac{\delta_1-\delta_2}{L_y}y\right]\cos(kx), \tag*{(2)}

where δ1(t)\delta_1(t) and δ2(t)\delta_2(t) correspond to the maximal deformations of the ceiling at ±Ly/2\pm L_y/2, and k=π/Lxk=\pi/L_x. Since the channel ceiling deformation scales linearly with the applied pressure [17], δ1\delta_1 and δ2\delta_2 are respectively assumed to be proportional to P1P_1 and P2P_2. The boundary conditions are chosen to enforce impermeability at the side walls (y=±Ly/2y=\pm L_y/2), and identical pressure values on the channel ends (x=±Lx/2x=\pm L_x/2). The potential ϕ0\phi_0 for the flow in the absence of obstacle can be derived analytically by solving Eq. (1) [21].

Assuming the obstacle is small compared to the domain (kR≪1kR\ll1), the velocity can be approximated as locally uniform on the scale of the obstacle [21]. The flow profile is then estimated using (U,V)=∇ϕ0(xc,yc)(U,V)=\nabla\phi_0(x_c,y_c), where (xc,yc)(x_c,y_c) is the obstacle center. The perturbed complex potential Φ1\Phi_1, and thus the flow around the obstacle, follows from the Milne-Thomson circle theorem [21, 23]. The streamlines predicted using the results of the potential model are plotted in the bottom row of Figure 2. Theoretical streamlines complete a full rotation around the obstacle over one period, and are in excellent agreement with experimental results.

We can now compute the trajectories of passive tracers in the flow. The time-varying Lagrangian velocity uL(x0,t)\mathbf{u}_{\mathrm{L}}(\mathbf{x}_0,t) of a tracer with initial position x0\mathbf{x}_0 in this unsteady flow is related to the local Eulerian velocity uE(x,t)\mathbf{u}_{\mathrm{E}}(\mathbf{x},t) through [14]:

uL(x0,t)=uE(x0+∫0tuL(x0,t′) dt′,t).(3)\mathbf{u}_{\mathrm{L}}(\mathbf{x}_0,t)=\mathbf{u}_{\mathrm{E}}\left(\mathbf{x}_0+\int_0^t\mathbf{u}_{\mathrm{L}}(\mathbf{x}_0,t')\,\mathrm{d}t',t\right). \tag*{(3)}

At first order in tracer excursion, the drift given by the potential flow solution vanishes over a period TT, i.e., ⟨uL⟩T≈⟨uE⟩T=0\langle u_{\mathrm{L}}\rangle_T \approx\langle u_{\mathrm{E}}\rangle_T = 0 [21]. At second order, a nonzero drift can be estimated [21]:

⟨uL⟩T≈2R4r5sin⁡(2kxc)k3h02Ly[1−cosh⁡(kyc)cosh⁡(kLy/2)]ATeθ,(4)\langle u_{\mathrm{L}}\rangle_T \approx\frac{2R^4}{r^5}\frac{\sin(2kx_c)}{k^3h_0^2L_y}\left[1-\frac{\cosh(ky_c)}{\cosh(kL_y/2)}\right]\frac{A}{T}\mathbf{e}_{\theta}, \tag*{(4)}

where rr is the distance to the center of the obstacle, and AA denotes the signed area enclosed by one cycle of the actuation loop in the (δ1,δ2)(\delta_1,\delta_2) plane. As observed in the Poincaré map of Figure 1(d), the net drift is purely azimuthal. Interestingly, this expression decouples the role of the actuation from the role of the obstacle and particle positions in the channel. Note for instance that two obstacles placed at streamwise positions symmetric with respect to the center of the chip will induce counter-rotating recirculations of equal intensity. In the specific case of sinusoidal actuation proportional to the profiles of P1P_1 and P2P_2 mentioned above [21]:

⟨uL⟩T≈α2R44r5sin⁡(2kxc)k3h02Ly[cosh⁡(kyc)cosh⁡(kLy/2)−1]ωPo2sin⁡φ eθ,(5)\langle u_{\mathrm{L}}\rangle_T \approx\frac{\alpha^2R^4}{4r^5}\frac{\sin(2kx_c)}{k^3h_0^2L_y}\left[\frac{\cosh(ky_c)}{\cosh(kL_y/2)}-1\right]\omega P_o^2\sin\varphi\,\mathbf{e}_{\theta}, \tag*{(5)}

where the constant α\alpha is defined such that αPi=−δi\alpha P_i=-\delta_i. Fluorescence height profilometry measurements yield α=0.5\alpha=0.5 µm/mbar. The net drift accumulates over successive cycles, as shown in the rightmost panels of Figure 2.

The angular displacement of fluid particles around the obstacle is determined experimentally using the computed trajectories of virtual particles seeded in an annulus of outer radius d=4Rd=4R around the obstacle, as shown in the inset of Figure 3(a). The angular displacement oscillates with a net overall drift that depends on the phase delay φ\varphi. The oscillations reflect the sinusoidal pressure actuation at period T=2T=2 s. When the actuation is in phase (φ=0\varphi=0), there is no net angular drift. When the actuation is asymmetric (φ=π/2\varphi=\pi/2 or 3π/23\pi/2), the drift is nonzero and its direction depends on the value of the phase delay φ\varphi, as shown in Figure 3(a).

Figure 3. (a) Azimuthal drift of particles around the obstacle. Inset: PIV results on an example Poincaré map. Red circles indicate the boundaries of the annular integration region (scale bar: 500 µm). (b) Average azimuthal velocity versus phase delay φ\varphi. Inset: Actuation loops in the (P1/Po,P2/Po)(P_1/P_o, P_2/P_o) plane for varying φ\varphi. (c) Average azimuthal velocity versus air chamber pressure PoP_o. Inset: Actuation loops in the (P1,P2)(P_1, P_2) plane for varying PoP_o. (d) Experimental versus theoretical average azimuthal velocity for 86 measurements, each one averaged over 30 periods. Unless varied, the experimental parameters were the following: φ=π2\varphi= \frac{\pi}{2}, ω=π\omega= \pi rad/s, xc=3.33x_c = 3.33 mm, yc=0y_c = 0 mm, R=375R = 375 µm, and Po=100P_o = 100 mbar. The gray dashed line indicates vth=vexpv_{\mathrm{th}} = v_{\mathrm{exp}}.

We then compute the experimental average (clockwise) azimuthal drift velocity vexpv_{\mathrm{exp}} in the same annulus for different values of the phase delay 0≤φ≤2π0\leq\varphi\leq2\pi, actuation frequency 0≤ω≤4π0\leq\omega\leq4\pi rad/s, and actuation amplitude 0≤Po≤2000\leq P_o\leq200 mbar, averaged over 30 periods. The experimental velocity shows a sinusoidal dependence on φ\varphi (Figure 3(b)), a linear dependence on the actuation frequency ω\omega (see Fig. S1 in [21]), and a quadratic dependence on the actuation amplitude PoP_o (Figure 3(c)), in agreement with the theoretical scalings of Eq. (5).

To compare experimental and theoretical results, the theoretical average azimuthal velocity vthv_{\mathrm{th}} is calculated in the same annulus around the obstacle using Eq. (5). The plot of vexpv_{\mathrm{exp}} as a function of vthv_{\mathrm{th}} shows that all data points collapse onto a single line (Figure 3(d)). This agreement is remarkable given that the model contains no adjustable parameters, and that the boundary conditions for the channel deformation represent a simplification of the actuation mechanism over the whole channel. Experiments and theory disagree most at frequencies ω>2π\omega>2\pi rad/s, which we attribute to the finite response time of the pneumatic control system to the actuating pressure.

Armed with this quantitative understanding of the recirculation, we now exploit this framework to create complex flow patterns in the device. Changing the shape of the obstacle modifies the recirculating pattern: for instance, a star-shaped obstacle still leads to a recirculation, albeit with a more complex pattern (Figure 4(a)). Similarly, combining multiple obstacles within the same channel leads to multiple recirculating patterns, which interact if the obstacles are close enough. Obstacles aligned along a diagonal line across the channel yield net transport along the axis defined by the arrangement of the obstacles (Figure 4(b), see also Movie S2 in [21]).

Four panels showing patterned flow fields around obstacles

Figure 4. Combination of patterned flow fields. (a) Poincaré map for the flow around a five-pointed star obstacle. (b) Poincaré map for the flow field resulting from the arrangement of four neighboring obstacles positioned diagonally with respect to the streamwise direction of the channel. (c), (d) Poincaré maps for the superposition of an externally controlled Poiseuille flow and the recirculation induced by the obstacle and the deformable ceiling, for φ=−π2\varphi= -\frac{\pi}{2} (c) or φ=π2\varphi= \frac{\pi}{2} (d). Scale bars: 500 µm.

Complex flow patterns can likewise emerge from combining actuation with an externally imposed flow, yielding a local flow router. We impose a constant flow rate Q=1Q = 1 µL/min in the microfluidic channel using a syringe pump (Pump 11 Elite, Harvard Apparatus). Without pneumatic actuation, the flow goes from left to right everywhere in the channel. For actuation at φ=−π2\varphi= -\frac{\pi}{2}, ω=π\omega= \pi rad/s, Po=100P_o = 100 mbar, the flow remains directed from left to right below the obstacle, but it is canceled by the recirculating flow above it (Figure 4(c), see Movie S3 in [21]). When the phase delay is reversed to φ=π2\varphi= \frac{\pi}{2}, the asymmetry is inverted (Figure 4(d), see Movie S3 in [21]). The fluid can thus be directed selectively to either side of the channel depending on the sign of sin⁡φ\sin\varphi.

In summary, the results above show that a static, passive obstacle in an oscillating Hele-Shaw cell rectifies a global, periodic wall deformation into a steady Lagrangian recirculation confined to its neighborhood. The net drift is due to purely kinematic effects, analogous to the well-known Stokes drift, and is well described by a potential flow model. This result recalls previous theoretical studies that predicted that periodically reoriented potential flows can serve to control Lagrangian trajectories and promote chaotic advection [24]. However, we now show that the presence of an obstacle in the domain confines the drift to its vicinity. This localization emerges from the strong local velocity gradients around the obstacle, which interact with the oscillating Eulerian field to generate the net drift through a second-order effect.

The model developed above shows that the displacement per cycle is set by a single geometric quantity: the signed area enclosed by the actuation cycle. A loop of zero area (e.g., φ=0\varphi= 0 or π\pi, a straight segment) produces no drift, which recalls the scallop theorem [25, 26, 27]. Varying φ\varphi and PoP_o (Figure 3(b) and Figure 3(c)) are therefore two different ways of varying the area AA enclosed by these loops, as shown in the insets of the corresponding panels. This dependence also predicts that the drift amplitude is independent of the shape of the waveform.

The flow structures constructed using this principle can be positioned, shaped, and combined either together or with an imposed external flow, making the topology of transport an accessible design variable at vanishing Reynolds number. Natural extensions of this work include mobile obstacles or additional actuators, which would offer further degrees of freedom for programmable flow control. Beyond the current experimental setup, the rectification of a periodically reoriented potential flow in the vicinity of an obstacle is a kinematic mechanism that does not depend on the underlying physical nature of the potential. It could therefore be extended to other transport problems modeled using the same equations, including flows in porous media or oscillating electric fields [8, 28].

Acknowledgments

The authors thank the X-Fab of École Polytechnique for its help and fruitful discussions. This work was funded by the European Union (ERC-2023-ADG grant number 101142018 MELCART). During the preparation of this work, the authors used Claude Opus 5 for code generation, assistance with model derivation, and improvement of the readability of the text. The authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Supplemental Material for Zero-Mean Oscillations Drive Recirculations in Potential Flows at Vanishing Reynolds Number

Alexandre S. Avaro; Leon V. Gebhard; Gabriel Amselem; Charles N. Baroud

supplemental_material.pdf

Supplemental Material for Zero-Mean Oscillations Drive Recirculations in Potential Flows at Vanishing Reynolds Number

Supplemental Movies

Movie S1. Flow around an example passive obstacle. The time-resolved frames show the instantaneous oscillatory flow. The stroboscopic sequence (i.e., sampled once per actuation cycle) reveals the net recirculating drift.

Movie S2. Flow around four obstacles arranged diagonally. The stroboscopic movie shows tracers both recirculating around each obstacle and drifting along the axis of the array.

Movie S3. Recirculation combined with an imposed pressure-driven flow. The stroboscopic movie shows the resulting drift under simultaneous air chamber actuation and pressure-driven flow applied using a syringe pump connected to the inlet of the channel. Reversing the actuation phase delay φ\varphi reverses the recirculation, moving the region of the domain where the net horizontal velocity vanishes.

Derivation of the drift expression

Derivation of the potential in the absence of obstacle

We first derive the potential solution ϕ0\phi_0 in the absence of obstacle. We consider a Hele-Shaw cell with dimensions LxL_x and LyL_y, such that x∈[−Lx/2,Lx/2]x \in[-L_x/2,L_x/2] and y∈[−Ly/2,Ly/2]y \in[-L_y/2,L_y/2]. Depth-averaged mass conservation yields:

∂h∂t+∇⋅(h⟨u⟩z)=0,(6)\frac{\partial h}{\partial t}+\nabla\cdot\left(h\langle u\rangle_z\right)=0, \tag*{(6)}

where h(x,y,t)h(x,y,t) is the local height of the channel and ⟨u⟩z\langle u\rangle_z is the depth-averaged velocity field. We define the ceiling deformation δ\delta such that h(x,y,t)=h0−δ(x,y,t)h(x,y,t)=h_0-\delta(x,y,t). Assuming δ/h0≪1\delta/h_0\ll1, the previous equation yields:

∇⋅⟨u⟩z=1h0∂δ∂t.(7)\nabla\cdot\langle u\rangle_z=\frac{1}{h_0}\frac{\partial\delta}{\partial t}. \tag*{(7)}

In the Hele-Shaw cell we can define a potential ϕ0\phi_0 such that ⟨u⟩z=∇ϕ0\langle u\rangle_z=\nabla\phi_0. This yields the following Poisson equation:

Δϕ0=1h0∂δ∂t.(8)\Delta\phi_0=\frac{1}{h_0}\frac{\partial\delta}{\partial t}. \tag*{(8)}

From the shape of the deformation shown in Figure 1 of the main text, we assume that the deformation profile is the following:

δ(x,y,t)=[δ1+δ22+δ1−δ2Lyy]cos⁡(kx),(9)\delta(x,y,t)=\left[\frac{\delta_1+\delta_2}{2}+\frac{\delta_1-\delta_2}{L_y}y\right]\cos(kx), \tag*{(9)}

where k=π/Lxk=\pi/L_x. This corresponds to a linear deformation in yy. Here, δ1\delta_1 and δ2\delta_2, once weighted in xx by cos⁡(kx)\cos(kx), correspond to the deformations on either side of the channel. We assume that δi\delta_i is proportional to PiP_i.

The problem then becomes:

Δϕ0=[Σ+Γy]cos⁡(kx),(10)\Delta\phi_0=[\Sigma+\Gamma y]\cos(kx), \tag*{(10)}

with

Σ=δ˙1+δ˙22h0,(11)\Sigma=\frac{\dot{\delta}_1+\dot{\delta}_2}{2h_0}, \tag*{(11)}
Γ=δ˙1−δ˙2h0Ly(12)\Gamma= \frac{\dot{\delta}_1-\dot{\delta}_2}{h_0L_y} \tag*{(12)}

We use the following boundary conditions:

∂ϕ0∂y(x,y=±Ly2,t)=0,(13)\frac{\partial\phi_0}{\partial y}\left(x,y=\pm\frac{L_y}{2},t\right)=0, \tag*{(13)}
ϕ0(x=±Lx2,y,t)=0.(14)\phi_0\left(x=\pm\frac{L_x}{2},y,t\right)=0. \tag*{(14)}

They respectively correspond to no flux through the side walls, and identical pressure at both ends of the channel. Using separation of variables, we look for solutions of the form ϕ0=f(y)cos⁡(kx)\phi_0=f(y)\cos(kx) (compatible with the boundary conditions at ±Lx/2\pm L_x/2). The problem then becomes the ordinary differential equation:

f′′−k2f=Σ+Γy.(15)f^{\prime\prime}-k^2f=\Sigma+\Gamma y. \tag*{(15)}

A particular solution is fp=−Σ+Γyk2f_p=-\frac{\Sigma+\Gamma y}{k^2}. The homogeneous solution is fh=C1cosh⁡(ky)+C2sinh⁡(ky)f_h=C_1\cosh(ky)+C_2\sinh(ky). The boundary conditions at ±Ly/2\pm L_y/2 yield:

−Γk2+C1ksinh⁡(kLy2)+C2kcosh⁡(kLy2)=0,(16)-\frac{\Gamma}{k^2}+C_1k\sinh\left(\frac{kL_y}{2}\right)+C_2k\cosh\left(\frac{kL_y}{2}\right)=0, \tag*{(16)}
−Γk2−C1ksinh⁡(kLy2)+C2kcosh⁡(kLy2)=0.(17)-\frac{\Gamma}{k^2}-C_1k\sinh\left(\frac{kL_y}{2}\right)+C_2k\cosh\left(\frac{kL_y}{2}\right)=0. \tag*{(17)}

These conditions entail C1=0C_1=0 and C2=Γk3cosh⁡(kLy/2)C_2=\frac{\Gamma}{k^3\cosh(kL_y/2)}. The potential solution in the absence of the obstacle is therefore:

ϕ0(x,y,t)=(Γk3sinh⁡(ky)cosh⁡(kLy/2)−Σ+Γyk2)cos⁡(kx).(18)\phi_0(x,y,t)=\left(\frac{\Gamma}{k^3}\frac{\sinh(ky)}{\cosh(kL_y/2)}-\frac{\Sigma+\Gamma y}{k^2}\right)\cos(kx). \tag*{(18)}
Perturbation description of the obstacle

We now treat the obstacle as a perturbation to the potential field. If the obstacle is smaller than the characteristic length of ϕ0\phi_0 (i.e., kR≪1kR\ll1), we can approximate the flow as locally uniform. We define the velocities at the center of the obstacle (xc,yc)(x_c,y_c):

U(t)=∂ϕ0∂x(xc,yc,t)=sin⁡(kxc)k[Σ+ΓG(yc)],(19)U(t)=\frac{\partial\phi_0}{\partial x}(x_c,y_c,t)=\frac{\sin(kx_c)}{k}\left[\Sigma+\Gamma G(y_c)\right], \tag*{(19)}
V(t)=∂ϕ0∂y(xc,yc,t)=cos⁡(kxc)H(yc)Γ,(20)V(t)=\frac{\partial\phi_0}{\partial y}(x_c,y_c,t)=\cos(kx_c)H(y_c)\Gamma, \tag*{(20)}

where:

G(yc)=yc−sinh⁡(kyc)kcosh⁡(kLy/2)(21)G(y_c)=y_c-\frac{\sinh(ky_c)}{k\cosh(kL_y/2)} \tag*{(21)}

and

H(yc)=1k2[cosh⁡(kyc)cosh⁡(kLy/2)−1].(22)H(y_c)=\frac{1}{k^2}\left[\frac{\cosh(ky_c)}{\cosh(kL_y/2)}-1\right]. \tag*{(22)}

The complex potential Φ1(z)\Phi_1(z) in the presence of the obstacle can be derived using the Milne-Thomson circle theorem:

Φ1(z)=U‾z+UR2z‾,(23)\Phi_1(z)=\overline{\mathcal{U}}z+\mathcal{U}\frac{R^2}{\overline{z}}, \tag*{(23)}

where U=U+iV\mathcal{U}=U+iV, RR is the radius of the obstacle, and zz the complex coordinate whose origin lies at the center of the obstacle. U‾\overline{\phantom{U}} denotes the complex conjugate. It follows that:

u−iv=U‾−UR2z‾ 2,(24)u-iv=\overline{\mathcal{U}}-\mathcal{U}\frac{R^2}{\overline{z}^{\,2}}, \tag*{(24)}

where uu and vv are the xx and yy components of the Eulerian velocity vector uE(x,t)\mathbf{u}_{\mathrm{E}}(x,t).

Mathematical interlude

We first define the following operations given a vector n\mathbf{n}:

n~(t)=∫0tn(t′) dt′,(25)\tilde{\mathbf{n}}(t)=\int_{0}^{t}\mathbf{n}(t')\,\mathrm{d}t', \tag*{(25)}
n˙=dndt,(26)\dot{\mathbf{n}}=\frac{\mathrm{d}\mathbf{n}}{\mathrm{d}t}, \tag*{(26)}
⟨n⟩T=1T∫0Tn(t) dt,(27)\langle\mathbf{n}\rangle_{T}=\frac{1}{T}\int_{0}^{T}\mathbf{n}(t)\,\mathrm{d}t, \tag*{(27)}
n=nx−iny.(28)\mathbf{n}=n_{x}-\mathrm{i}n_{y}. \tag*{(28)}

Here, nxn_{x} and nyn_{y} are the projections of n\mathbf{n} on the xx and yy directions. Note that n\mathbf{n} is a complex number, and that the real and imaginary parts of nˉ=nx+iny\bar{\mathbf{n}}=n_{x}+\mathrm{i}n_{y} correspond to the xx and yy components of n\mathbf{n}. We note the following properties for integrable TT-periodic functions n1(t)n_{1}(t) and n2(t)n_{2}(t) such that ⟨n1⟩T=⟨n2⟩T=0\langle n_{1}\rangle_{T}=\langle n_{2}\rangle_{T}=0:

⟨n1n~1⟩T=0,(29)\langle n_{1}\tilde{n}_{1}\rangle_{T}=0, \tag*{(29)}
⟨n1n~2⟩T=−⟨n~1n2⟩T.(30)\langle n_{1}\tilde{n}_{2}\rangle_{T}=-\langle\tilde{n}_{1}n_{2}\rangle_{T}. \tag*{(30)}

In particular, because δ1\delta_{1} and δ2\delta_{2} are TT-periodic, δ˙1\dot{\delta}_{1} and δ˙2\dot{\delta}_{2} are also TT-periodic and ⟨δ˙1⟩T=⟨δ˙2⟩T=0\langle\dot{\delta}_{1}\rangle_{T}=\langle\dot{\delta}_{2}\rangle_{T}=0. It follows from this that ⟨Σ⟩T=⟨Γ⟩T=0\langle\Sigma\rangle_{T}=\langle\Gamma\rangle_{T}=0, and therefore ⟨U⟩T=⟨V⟩T=0\langle U\rangle_{T}=\langle V\rangle_{T}=0, and ⟨uE⟩T=0\langle u_{\mathrm{E}}\rangle_{T}=0. Hence, properties (S24) and (S25) hold for δ˙1\dot{\delta}_{1}, δ˙2\dot{\delta}_{2}, Σ\Sigma, Γ\Gamma, UU, and VV. Another useful property is the following:

(n1⋅∇)n2=nˉ1 n2′.(31)(\mathbf{n}_{1}\cdot\nabla)\mathbf{n}_{2}=\bar{n}_{1}\,n'_{2}. \tag*{(31)}
Computation of the drift over one period

The drift of a particle with initial position x0\mathbf{x}_{0} is the integral of the Lagrangian velocity uL(x0,t)\mathbf{u}_{\mathrm{L}}(\mathbf{x}_{0},t) over an actuation period. uL\mathbf{u}_{\mathrm{L}} is given by:

uL(x0,t)=uE(x0+u~E,t).(32)\mathbf{u}_{\mathrm{L}}(\mathbf{x}_{0},t)=\mathbf{u}_{\mathrm{E}}(\mathbf{x}_{0}+\tilde{\mathbf{u}}_{\mathrm{E}},t). \tag*{(32)}

At first order, the drift given by the potential flow solution vanishes over a period TT:

⟨uL⟩T≈⟨uE⟩T=0.(33)\langle\mathbf{u}_{\mathrm{L}}\rangle_{T}\approx\langle\mathbf{u}_{\mathrm{E}}\rangle_{T}=0. \tag*{(33)}

At second order, the drift is:

⟨uL⟩T≈⟨u~E⋅∇uE⟩T.(34)\langle\mathbf{u}_{\mathrm{L}}\rangle_{T}\approx\langle\tilde{\mathbf{u}}_{\mathrm{E}}\cdot\nabla\mathbf{u}_{\mathrm{E}}\rangle_{T}. \tag*{(34)}

Because uu and vv are linear combinations of UU and VV, there exist vectors a\mathbf{a} and b\mathbf{b} such that:

uE(x,t)=U(t)a(x)+V(t)b(x).(35)\mathbf{u}_{\mathrm{E}}(\mathbf{x},t)=U(t)\mathbf{a}(\mathbf{x})+V(t)\mathbf{b}(\mathbf{x}). \tag*{(35)}

The drift is therefore given by:

⟨u~E⋅∇uE⟩T=⟨U~U⟩T(a⋅∇)a+⟨V~V⟩T(b⋅∇)b+⟨U~V⟩T(a⋅∇)b+⟨V~U⟩T(b⋅∇)a.(36)\langle\tilde{\mathbf{u}}_{\mathrm{E}}\cdot\nabla\mathbf{u}_{\mathrm{E}}\rangle_{T} =\langle\tilde{U}U\rangle_{T}(\mathbf{a}\cdot\nabla)\mathbf{a} +\langle\tilde{V}V\rangle_{T}(\mathbf{b}\cdot\nabla)\mathbf{b} +\langle\tilde{U}V\rangle_{T}(\mathbf{a}\cdot\nabla)\mathbf{b} +\langle\tilde{V}U\rangle_{T}(\mathbf{b}\cdot\nabla)\mathbf{a}. \tag*{(36)}

Using properties (S24) and (S25), this reduces to:

⟨u~E⋅∇uE⟩T=⟨U~V⟩TK,(37)\langle\tilde{\mathbf{u}}_{\mathrm{E}}\cdot\nabla\mathbf{u}_{\mathrm{E}}\rangle_{T} =\langle\tilde{U}V\rangle_{T}\mathbf{K}, \tag*{(37)}

where:

K=(a⋅∇)b−(b⋅∇)a.(38)\mathbf{K}=(\mathbf{a}\cdot\nabla)\mathbf{b}-(\mathbf{b}\cdot\nabla)\mathbf{a}. \tag*{(38)}

We first evaluate ⟨U~V⟩T\langle\widetilde{U}V\rangle_T. From their respective definitions and Eq. (29):

⟨U~V⟩T=sin⁡(kxc)cos⁡(kxc)kH(yc)[⟨Σ~+Γ~G(yc)⟩T]Γ=sin⁡(2kxc)2kH(yc)⟨Σ~Γ⟩T.(39)\langle\widetilde{U}V\rangle_T=\frac{\sin(kx_c)\cos(kx_c)}{k}H(y_c)\left[\langle\widetilde{\Sigma}+\widetilde{\Gamma}G(y_c)\rangle_T\right]\Gamma=\frac{\sin(2kx_c)}{2k}H(y_c)\langle\widetilde{\Sigma}\Gamma\rangle_T. \tag*{(39)}

Substituting Σ\Sigma and Γ\Gamma from their respective definitions:

⟨U~V⟩T=sin⁡(2kxc)2k3h02Ly[1−cosh⁡(kyc)cosh⁡(kLy/2)]⟨δ1δ˙2⟩T.(40)\langle\widetilde{U}V\rangle_T=\frac{\sin(2kx_c)}{2k^3h_0^2L_y}\left[1-\frac{\cosh(ky_c)}{\cosh(kL_y/2)}\right]\langle\delta_1\dot{\delta}_2\rangle_T. \tag*{(40)}

We now compute K\mathbf{K}. From the chosen flow profile (Eq. (24)) and the definitions of a\mathbf{a} and b\mathbf{b} (Eq. (35)):

uE=U‾−uR2z2=U(t)a+V(t)b.(41)\mathbf{u}_E=\overline{U}-u\frac{R^2}{z^2}=U(t)\mathbf{a}+V(t)\mathbf{b}. \tag*{(41)}

Because uE\mathbf{u}_E is linear in UU and VV, we identify:

a=1−R2z2,(42)\mathbf{a}=1-\frac{R^2}{z^2}, \tag*{(42)}
b=−i(1+R2z2).(43)\mathbf{b}=-i\left(1+\frac{R^2}{z^2}\right). \tag*{(43)}

Then, we note using (S26) that:

K=a‾b′−b‾a′.(44)\mathbf{K}=\overline{\mathbf{a}}\mathbf{b}'-\overline{\mathbf{b}}\mathbf{a}'. \tag*{(44)}

This can be easily computed using the previous result:

a‾b′=2iR2z3−2iR4z3z‾2,(45)\overline{\mathbf{a}}\mathbf{b}'=\frac{2iR^2}{z^3}-\frac{2iR^4}{z^3\overline{z}^2}, \tag*{(45)}
b‾a′=2iR2z3+2iR4z3z‾2,(46)\overline{\mathbf{b}}\mathbf{a}'=\frac{2iR^2}{z^3}+\frac{2iR^4}{z^3\overline{z}^2}, \tag*{(46)}

so that:

K=−4iR4z3z‾2,(47)\mathbf{K}=-\frac{4iR^4}{z^3\overline{z}^2}, \tag*{(47)}

and

K‾=4iR4z2z‾3.(48)\overline{\mathbf{K}}=\frac{4iR^4}{z^2\overline{z}^3}. \tag*{(48)}

Using the polar coordinates (r,θ)(r,\theta) centered at the obstacle such that z=reiθz=re^{i\theta}:

K‾=4R4r5ieiθ.(49)\overline{\mathbf{K}}=\frac{4R^4}{r^5}ie^{i\theta}. \tag*{(49)}

By identification of the real and imaginary parts of K‾\overline{\mathbf{K}} we find the xx and yy components of K\mathbf{K}. Therefore:

K=4R4r5(−sin⁡θ ex+cos⁡θ ey)=4R4r5eθ,(50)\mathbf{K}=\frac{4R^4}{r^5}(-\sin\theta\,\mathbf{e}_x+\cos\theta\,\mathbf{e}_y)=\frac{4R^4}{r^5}\mathbf{e}_\theta, \tag*{(50)}

where ex\mathbf{e}_x, ey\mathbf{e}_y, and eθ\mathbf{e}_\theta are the unit vectors in the xx, yy, and θ\theta directions. The drift can then be written as:

⟨u~E⋅∇uE⟩T=2R4r5sin⁡(2kxc)k3h02Ly[1−cosh⁡(kyc)cosh⁡(kLy/2)]⟨δ1δ˙2⟩Teθ.(51)\langle\widetilde{\mathbf{u}}_E\cdot\nabla\mathbf{u}_E\rangle_T=\frac{2R^4}{r^5}\frac{\sin(2kx_c)}{k^3h_0^2L_y}\left[1-\frac{\cosh(ky_c)}{\cosh(kL_y/2)}\right]\langle\delta_1\dot{\delta}_2\rangle_T\mathbf{e}_\theta. \tag*{(51)}

Finally, we can note that ⟨δ1δ˙2⟩T\langle\delta_1\dot{\delta}_2\rangle_T is related to the signed area A\mathcal{A} enclosed by the actuation loop in the (δ1,δ2)(\delta_1,\delta_2) space. Indeed:

⟨δ1δ˙2⟩T=1T∫0Tδ1δ˙2 dt=1T∮δ1 dδ2=AT.(52)\langle\delta_1\dot{\delta}_2\rangle_T=\frac{1}{T}\int_0^T\delta_1\dot{\delta}_2\,\mathrm{d}t=\frac{1}{T}\oint\delta_1\,\mathrm{d}\delta_2=\frac{\mathcal{A}}{T}. \tag*{(52)}

This yields the final result:

⟨u~E⋅∇uE⟩T=2R4r5sin⁡(2kxc)k3h02Ly[1−cosh⁡(kyc)cosh⁡(kLy/2)]ATeθ.(53)\langle\tilde{\mathbf{u}}_E\cdot\nabla\mathbf{u}_E\rangle_T=\frac{2R^4}{r^5}\frac{\sin(2kx_c)}{k^3h_0^2L_y}\left[1-\frac{\cosh(ky_c)}{\cosh(kL_y/2)}\right]\frac{\mathcal{A}}{T}\mathbf{e}_{\theta}. \tag*{(53)}

In the case of sinusoidal actuation of equal amplitude, A\mathcal{A} is the area of a Lissajous ellipse. Indeed, if the ceiling deformation is δi=δ02(1+sin⁡(2πtT+{0,φ}))\delta_i=\frac{\delta_0}{2}\left(1+\sin\left(\frac{2\pi t}{T}+\{0,\varphi\}\right)\right), then:

A=πδ022T∫0Tsin⁡(2πtT)cos⁡(2πtT+φ) dt=−π4δ02sin⁡φ,(54)\mathcal{A}=\frac{\pi\delta_0^2}{2T}\int_0^T\sin\left(\frac{2\pi t}{T}\right)\cos\left(\frac{2\pi t}{T}+\varphi\right)\,\mathrm{d}t=-\frac{\pi}{4}\delta_0^2\sin\varphi, \tag*{(54)}

and:

⟨u~E⋅∇uE⟩T=R44r5sin⁡(2kxc)k3h02Ly[cosh⁡(kyc)cosh⁡(kLy/2)−1]ωδ02sin⁡φ eθ.(55)\langle\tilde{\mathbf{u}}_E\cdot\nabla\mathbf{u}_E\rangle_T=\frac{R^4}{4r^5}\frac{\sin(2kx_c)}{k^3h_0^2L_y}\left[\frac{\cosh(ky_c)}{\cosh(kL_y/2)}-1\right]\omega\delta_0^2\sin\varphi\,\mathbf{e}_{\theta}. \tag*{(55)}

Velocity measurements for varying actuation frequency

Figure S1. Average azimuthal drift velocity vexpv_{\mathrm{exp}} versus actuation frequency ω\omega. Other experimental parameters are identical to those reported for Figure 3 of the main text.

References

  1. [1]E. E. Kriezis, T. D. Tsiboukis, S. M. Panas, and J. A. Tegopoulos, Eddy currents: Theory and applications, Proceedings of the IEEE 80, 1559 (1992).DOI
  2. [2]J. O. Dabiri, S. P. Colin, J. H. Costello, and M. Gharib, Flow patterns generated by oblate medusan jellyfish: Field measurements and laboratory analyses, Journal of Experimental Biology 208, 1257 (2005).pubmed.ncbi.nlm.nih.gov/15781886
  3. [3]J. Sommeria, S. D. Meyers, and H. L. Swinney, Laboratory simulation of Jupiter’s Great Red Spot, Nature 331, 689 (1988).DOI
  4. [4]H. K. Moffatt, Viscous and resistive eddies near a sharp corner, Journal of Fluid Mechanics 18, 1 (1964).
  5. [5]S. Jakiela, P. M. Korczyk, S. Makulska, O. Cybulski, and P. Garstecki, Discontinuous Transition in a Laminar Fluid Flow: A Change of Flow Topology inside a Droplet Moving in a Micron-Size Channel, Physical Review Letters 108, 134501 (2012).pubmed.ncbi.nlm.nih.gov/22540703
  6. [6]R. L. Panton, Incompressible Flow, 4th ed. (John Wiley & Sons, 2013).
  7. [7]F. Gallaire, P. Meliga, P. Laure, and C. N. Baroud, Marangoni induced force on a drop in a Hele Shaw cell, Physics of Fluids 26, 062105 (2014).DOI
  8. [8]K. I. McKee and J. W. M. Bush, Potential flows with electromagnetically induced circulation in a Hele-Shaw cell, Physical Review Fluids 10, 054103 (2025).arxiv.org/abs/2410.07366
  9. [9]B. Rallabandi, C. Wang, and S. Hilgenfeldt, Two-dimensional streaming flows driven by sessile semicylindrical microbubbles, Journal of Fluid Mechanics 739, 57 (2014).DOI
  10. [10]X. Zhang and B. Rallabandi, Three-dimensional streaming around an obstacle in a Hele-Shaw cell, Journal of Fluid Mechanics 961, A35 (2023).DOI
  11. [11]K. Loutherback, J. Puchalla, R. H. Austin, and J. C. Sturm, Deterministic Microfluidic Ratchet, Physical Review Letters 102, 045301 (2009).pubmed.ncbi.nlm.nih.gov/19257437
  12. [12]C. Kurzthaler, D. L. Chase, and H. A. Stone, Surface corrugations induce helical near-surface flows and transport in microfluidic channels, Journal of Fluid Mechanics 982, A31 (2024).DOI
  13. [13]B. Nordén, Y. Zolotaryuk, P. L. Christiansen, and A. V. Zolotaryuk, Ratchet device with broken friction symmetry, Applied Physics Letters 80, 2601 (2002).DOI
  14. [14]G. G. Stokes, On the theory of oscillatory waves, Transactions of the Cambridge Philosophical Society 8, 441 (1847).
  15. [15]M. S. Longuet-Higgins, Steady currents induced by oscillations round islands, Journal of Fluid Mechanics 42, 701 (1970).
  16. [16]T. S. van den Bremer and Ø. Breivik, Stokes drift, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 376, 20170104 (2018).DOI
  17. [17]L. V. Gebhard, A. S. Avaro, G. Amselem, and C. N. Baroud, Designing single-layer PDMS devices for micron to millimeter-scale deformations (2026), arXiv:2605.17402 [physics.flu-dyn].arxiv.org/abs/2605.17402
  18. [18]S. Jain, H. Belkadi, A. Michaut, S. Sart, J. Gros, M. Genet, and C. N. Baroud, Using a micro-device with a deformable ceiling to probe stiffness heterogeneities within 3D cell aggregates, Biofabrication 16, 035010 (2024).DOI
  19. [19]R. Dangla, S. Lee, and C. N. Baroud, Trapping Microfluidic Drops in Wells of Surface Energy, Physical Review Letters 107, 124501 (2011).pubmed.ncbi.nlm.nih.gov/22026771
  20. [20]M. C. M. Sequeira, H. M. N. T. Avelino, F. J. P. Caetano, and J. M. N. A. Fareleira, Viscosity and Density Measurements of Poly(ethyleneglycol) 200 and Poly(ethyleneglycol) 600 at High Pressures, Journal of Chemical & Engineering Data 68, 64 (2023).DOI
  21. [21]See Supplemental Material at [URL] for the derivation of the flow potential and drift, Fig. S1, and Movies S1, S2, and S3.
  22. [22]W. Thielicke and E. J. Stamhuis, PIVLab – Towards User-friendly, Affordable and Accurate Digital Particle Image Velocimetry in MATLAB, Journal of Open Research Software 2, e30 (2014).DOI
  23. [23]L. M. Milne-Thomson, Theoretical Hydrodynamics, 4th ed. (Macmillan, 1960).
  24. [24]D. R. Lester, G. Metcalfe, M. G. Trefry, A. Ord, B. Hobbs, and M. Rudman, Lagrangian topology of a periodically reoriented potential flow: Symmetry, optimization, and mixing, Physical Review E 80, 036208 (2009).pubmed.ncbi.nlm.nih.gov/19905201
  25. [25]E. M. Purcell, Life at low Reynolds number, American Journal of Physics 45, 3 (1977).
  26. [26]A. Shapere and F. Wilczek, Geometry of self-propulsion at low Reynolds number, Journal of Fluid Mechanics 198, 557 (1989).DOI
  27. [27]G. Amselem, C. Clanet, and M. Benzaken, Valveless Pumping at Low Reynolds Numbers, Physical Review Applied 19, 024017 (2023).DOI
  28. [28]F. Box, G. Peng, D. Pihler-Puzović, and A. Juel, Flow-induced choking of a compliant Hele-Shaw cell, Proceedings of the National Academy of Sciences 117, 30228 (2020).DOI

Paper details

vlnc:W_bk3672rv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Claude Opus 5 (bare)

View statement
Looking for related papers…
Contents