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 , much smaller than its width and length . 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 . 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 m.
Imposing different negative pressures in each air chamber yields an asymmetric channel ceiling deformation. We concentrate on periodic actuations. A pressure is imposed in one air chamber, and a pressure in the other. Here, and respectively denote the amplitude and angular frequency of the actuation, and the phase delay between the two actuations. The time evolution of the deformation of the entire microfluidic channel for and mbar is shown in Figure 1(b). The deformation profile is symmetric along when both actuating pressures are equal, and tilted otherwise, with a maximum deformation . Corresponding transverse channel height profiles along 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 and .
The microfluidic channel includes five small traps etched in the ceiling of the device and centered along [19]. These traps appear as darker dots in Figure 1(b), and as a kink at in the profiles shown in Figure 1(c). A fixed obstacle, here a cylindrical hydrogel of radius , 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 [20], and seeded with fluorescent particles of diameter 3 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 (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 (where is a characteristic instantaneous velocity) and the Womersley number are very small. We first define a channel without obstacle, of dimensions and . The Eulerian flow field satisfies , where the velocity potential obeys the Poisson equation whose source term represents the effects of vertical ceiling actuation throughout the domain:
Here, is the ceiling displacement. Experimentally, the ceiling deformation is not constant along , but shows a maximum in the center of the device () and a minimum at the device edges (), which is modeled as a cosine profile (Figure 1(b)). Further, we approximate as a linear function of (Figure 1(c)). is therefore given by:
where and correspond to the maximal deformations of the ceiling at , and . Since the channel ceiling deformation scales linearly with the applied pressure [17], and are respectively assumed to be proportional to and . The boundary conditions are chosen to enforce impermeability at the side walls (), and identical pressure values on the channel ends (). The potential 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 (), the velocity can be approximated as locally uniform on the scale of the obstacle [21]. The flow profile is then estimated using , where is the obstacle center. The perturbed complex potential , 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 of a tracer with initial position in this unsteady flow is related to the local Eulerian velocity through [14]:
At first order in tracer excursion, the drift given by the potential flow solution vanishes over a period , i.e., [21]. At second order, a nonzero drift can be estimated [21]:
where is the distance to the center of the obstacle, and denotes the signed area enclosed by one cycle of the actuation loop in the 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 and mentioned above [21]:
where the constant is defined such that . Fluorescence height profilometry measurements yield 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 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 . The oscillations reflect the sinusoidal pressure actuation at period s. When the actuation is in phase (), there is no net angular drift. When the actuation is asymmetric ( or ), the drift is nonzero and its direction depends on the value of the phase delay , 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 . Inset: Actuation loops in the plane for varying . (c) Average azimuthal velocity versus air chamber pressure . Inset: Actuation loops in the plane for varying . (d) Experimental versus theoretical average azimuthal velocity for 86 measurements, each one averaged over 30 periods. Unless varied, the experimental parameters were the following: , rad/s, mm, mm, µm, and mbar. The gray dashed line indicates .
We then compute the experimental average (clockwise) azimuthal drift velocity in the same annulus for different values of the phase delay , actuation frequency rad/s, and actuation amplitude mbar, averaged over 30 periods. The experimental velocity shows a sinusoidal dependence on (Figure 3(b)), a linear dependence on the actuation frequency (see Fig. S1 in [21]), and a quadratic dependence on the actuation amplitude (Figure 3(c)), in agreement with the theoretical scalings of Eq. (5).
To compare experimental and theoretical results, the theoretical average azimuthal velocity is calculated in the same annulus around the obstacle using Eq. (5). The plot of as a function of 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 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]).

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 (c) or (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 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 , rad/s, 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 , 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 .
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., or , a straight segment) produces no drift, which recalls the scallop theorem [25, 26, 27]. Varying and (Figure 3(b) and Figure 3(c)) are therefore two different ways of varying the area 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 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 in the absence of obstacle. We consider a Hele-Shaw cell with dimensions and , such that and . Depth-averaged mass conservation yields:
where is the local height of the channel and is the depth-averaged velocity field. We define the ceiling deformation such that . Assuming , the previous equation yields:
In the Hele-Shaw cell we can define a potential such that . This yields the following Poisson equation:
From the shape of the deformation shown in Figure 1 of the main text, we assume that the deformation profile is the following:
where . This corresponds to a linear deformation in . Here, and , once weighted in by , correspond to the deformations on either side of the channel. We assume that is proportional to .
The problem then becomes:
with
We use the following boundary conditions:
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 (compatible with the boundary conditions at ). The problem then becomes the ordinary differential equation:
A particular solution is . The homogeneous solution is . The boundary conditions at yield:
These conditions entail and . The potential solution in the absence of the obstacle is therefore:
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 (i.e., ), we can approximate the flow as locally uniform. We define the velocities at the center of the obstacle :
where:
and
The complex potential in the presence of the obstacle can be derived using the Milne-Thomson circle theorem:
where , is the radius of the obstacle, and the complex coordinate whose origin lies at the center of the obstacle. denotes the complex conjugate. It follows that:
where and are the and components of the Eulerian velocity vector .
Mathematical interlude
We first define the following operations given a vector :
Here, and are the projections of on the and directions. Note that is a complex number, and that the real and imaginary parts of correspond to the and components of . We note the following properties for integrable -periodic functions and such that :
In particular, because and are -periodic, and are also -periodic and . It follows from this that , and therefore , and . Hence, properties (S24) and (S25) hold for , , , , , and . Another useful property is the following:
Computation of the drift over one period
The drift of a particle with initial position is the integral of the Lagrangian velocity over an actuation period. is given by:
At first order, the drift given by the potential flow solution vanishes over a period :
At second order, the drift is:
Because and are linear combinations of and , there exist vectors and such that:
The drift is therefore given by:
Using properties (S24) and (S25), this reduces to:
where:
We first evaluate . From their respective definitions and Eq. (29):
Substituting and from their respective definitions:
We now compute . From the chosen flow profile (Eq. (24)) and the definitions of and (Eq. (35)):
Because is linear in and , we identify:
Then, we note using (S26) that:
This can be easily computed using the previous result:
so that:
and
Using the polar coordinates centered at the obstacle such that :
By identification of the real and imaginary parts of we find the and components of . Therefore:
where , , and are the unit vectors in the , , and directions. The drift can then be written as:
Finally, we can note that is related to the signed area enclosed by the actuation loop in the space. Indeed:
This yields the final result:
In the case of sinusoidal actuation of equal amplitude, is the area of a Lissajous ellipse. Indeed, if the ceiling deformation is , then:
and:
Velocity measurements for varying actuation frequency
Figure S1. Average azimuthal drift velocity versus actuation frequency . Other experimental parameters are identical to those reported for Figure 3 of the main text.
References
- [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]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]J. Sommeria, S. D. Meyers, and H. L. Swinney, Laboratory simulation of Jupiter’s Great Red Spot, Nature 331, 689 (1988).DOI
- [4]H. K. Moffatt, Viscous and resistive eddies near a sharp corner, Journal of Fluid Mechanics 18, 1 (1964).
- [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]R. L. Panton, Incompressible Flow, 4th ed. (John Wiley & Sons, 2013).
- [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]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]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]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]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]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]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]G. G. Stokes, On the theory of oscillatory waves, Transactions of the Cambridge Philosophical Society 8, 441 (1847).
- [15]M. S. Longuet-Higgins, Steady currents induced by oscillations round islands, Journal of Fluid Mechanics 42, 701 (1970).
- [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]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]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]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]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]See Supplemental Material at [URL] for the derivation of the flow potential and drift, Fig. S1, and Movies S1, S2, and S3.
- [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]L. M. Milne-Thomson, Theoretical Hydrodynamics, 4th ed. (Macmillan, 1960).
- [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]E. M. Purcell, Life at low Reynolds number, American Journal of Physics 45, 3 (1977).
- [26]A. Shapere and F. Wilczek, Geometry of self-propulsion at low Reynolds number, Journal of Fluid Mechanics 198, 557 (1989).DOI
- [27]G. Amselem, C. Clanet, and M. Benzaken, Valveless Pumping at Low Reynolds Numbers, Physical Review Applied 19, 024017 (2023).DOI
- [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



