Hydrodynamic schooling of flapping swimmers

Abstract

Fish schools and bird flocks are fascinating examples of collective behaviours in which many individuals generate and interact with complex flows. Motivated by animal groups on the move, here we explore how the locomotion of many bodies emerges from their flow-mediated interactions. Through experiments and simulations of arrays of flapping wings that propel within a collective wake, we discover distinct modes characterized by the group swimming speed and the spatial phase shift between trajectories of neighbouring wings. For identical flapping motions, slow and fast modes coexist and correspond to constructive and destructive wing-wake interactions. Simulations show that swimming in a group can enhance speed and save power, and we capture the key phenomena in a mathematical model based on memory or the storage and recollection of information in the flow field. These results also show that fluid dynamic interactions alone are sufficient to generate coherent collective locomotion, and thus might suggest new ways to characterize the role of flows in animal groups.

ARTICLE

Received 8 May 2015 | Accepted 2 Sep 2015 | Published 6 Oct 2015

Fish schools and bird ocks are fascinating examples of collective behaviours in which many individuals generate and interact with complex ows. Motivated by animal groups on the move, here we explore how the locomotion of many bodies emerges from their ow-mediated interactions. Through experiments and simulations of arrays of apping wings that propel within a collective wake, we discover distinct modes characterized by the group swimming speed and the spatial phase shift between trajectories of neighbouring wings. For identical apping motions, slow and fast modes coexist and correspond to constructive and destructive wingwake interactions. Simulations show that swimming in a group can enhance speed and save power, and we capture the key phenomena in a mathematical model based on memory or the storage and recollection of information in the ow eld. These results also show that uid dynamic interactions alone are sufcient to generate coherent collective locomotion, and thus might suggest new ways to characterize the role of ows in animal groups.

DOI: 10.1038/ncomms9514 OPEN

Hydrodynamic schooling of apping swimmers

Alexander D. Becker1,*, Hassan Masoud1,*,w, Joel W. Newbolt1, Michael Shelley1 & Leif Ristroph1

1 Applied Math Lab, Courant Institute, New York University, 251 Mercer Street, New York, New York 10012, USA. * These authors contributed equally to this work. w Present address: Department of Mechanical Engineering, University of Nevada, Reno, Reno, Nevada 89557, USA. Correspondence and requests for materials should be addressed to L.R. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514

a

When collections of bodies move within a uid, the motion of each is inuenced by the ows generated by others, often with surprising and important

consequences. During sedimentation of particulate suspensions, for example, many-body uid-mediated interactions lead to unusual particle trajectories as well as large-scale ows that signicantly affect the global settling time1. Similar interactions are at work in many natural contexts and engineering applications, from water droplets within clouds2 to beds of sand or other granular matter suspended by ows3. These cases in which bodies passively respond to external forcing are complemented by systems in which the constituents actively generate the ows through which they interact46. To date, studies of such active suspensions or active matter have largely focused on swimming micro-organisms, or micro-particles, that produce ows through chemical reactions79. At such small scales, viscosity dominates inertia, ow elds are established nearly instantaneously, and thus interactions between bodies can be thought of as immediate and depending only on the present conguration and motions10.

For larger bodies and faster motions where uid inertia is important, ows decay slowly, and interactions can no longer be viewed as instantaneous. The biological realm provides some of the most fascinating examples in which individuals actively generate inertial ows, such as schooling of sh and ocking of birds. The collective behaviours are intriguing from many perspectives1114, including the conventional uid dynamic view that highly ordered animal groupings benet from ow-mediated interactions by saving on the energetic cost of movement15,16. For example, V-formation ight of birds is thought to involve favourable interactions with the up- and down-wash of upstream neighbours1618, and correlations in apping motions and inter-neighbour spacing in actual ocks have been interpreted as behavioural responses that take advantage of these ows19,20. For sh schools, a diamond-shaped lattice arrangement is thought to enable individuals to extract energy from the vortices shed by others15,21, although eld and laboratory studies have led to conicting conclusions regarding a possible hydrodynamic function of schooling22,23.

Inspired by animal groupings, here we seek to understand how uid-mediated interactions among self-propelling bodies inuence their collective locomotion. Specically, we study such interactions in the idealized setting of linear arrays of wings undergoing prescribed apping motions but whose swimming speed is free or dynamically determined. Because the ows are inertial or long-lived, the forcing experienced by each body depends on the dynamical history of the ensemble, and the system can be said to possess memory of past states. We aim to understand how this effect is manifest in collective locomotion and to draw parallels to other uid-structure systems that display memory2426. Understanding these interactions could also provide insight into the uid dynamics of animal groups on the move, especially with regard to possible exploitation of ows for formation locomotion or highly ordered groupings. Moreover, the underlying hydrodynamic principles might be put to use in applications involving the interactions of bodies with unsteady ows2732. For example, schemes might be devised for harvesting energy from waves or other time-dependent ows, and air or water vehicles could be designed to take advantage of interactions between propulsors.

ResultsExperimental approach. Our experimental design is guided by the goals of achieving self-propulsion for prescribed apping motions and realizing an effectively innite array of bodies to

explore their interactions. As shown in Fig. 1a and detailed in the Methods section, the device consists of one or more wings or hydrofoils that are heaved up and down, and consequently swim in orbits around a cylindrical water tank3335. A video showing the device in operation is available in Supplementary Movie 1. Free swimming is achieved by mounting the wings to a vertical axle via a low-friction rotary bearing. Thus, the forward locomotion is not prescribed but is an outcome of the interaction with the uid. This self-propulsion condition as well as the inertial or high Reynolds number ows are important properties shared with bird ight and sh swimming. Our system, however, involves xed separations between bodies, whereas inter-individual spacing is free or dynamically determined in animal groups, and this difference will likely lead to differences in the emergent locomotion.

Previous studies have shown that the dynamics and ows observed in rotational systems compare well with those in translational geometries3336. To ensure the phenomena we observe are not a peculiarity of geometry, we accompany our three-dimensional (3D) rotational experiments with computational simulations in two-dimensional (2D) translational domains, as discussed in detail below. Most importantly, the rotational geometry induces each wing to swim within the wakes of others, as indicated by the trajectories of Fig. 1b, and this arrangement thus mimics the conditions in an innite array. The device also allows us to explore swimming behaviour for varying strength of hydrodynamic interactions between swimmers. Increasing the number of wings connected to the assembly decreases the spacing between members in an array and amplies interactions. Further, increasing apping frequency or amplitude induces faster swimming and thus stronger ows to

Motor assembly & rotary bearing

Encoder

Water tank

F

f

Wing assembly

b

1st Pass 2nd Pass

Figure 1 | Flapping wings swimming in rotational orbits mimic an innite array of locomotors. (a) A motor heaves a wing or wing pair up and down at prescribed frequency f and peak-to-peak amplitude A, resulting in swimming of rotational frequency F around a cylindrical water tank. (b) The rotational geometry allows for interactions with the ows generated in previous orbits.

2 NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514 ARTICLE

be encountered in subsequent orbits. The swimming dynamics are characterized by measuring the rotational speed around the tank, and use of a clear-walled tank allows for visualization of the ow elds.

Dynamics of interacting wings. Previous studies have used rotational systems to study the dynamics and hydrodynamics relevant to an isolated wing3335. This is achieved by relatively slow swimming of widely spaced hydrofoils, in which case interactions are weak and the swimming speed follows simple dependencies on apping frequency and amplitude. Here we aim to induce strong interactions by considering fast swimming of a wing pair, and indeed we observe markedly different dynamics. To systematically characterize the locomotion, we vary both the peak-to-peak amplitude A and apping frequency f for the pair and measure the resulting rotational frequency F around the tank. For each A, we incrementally increase f, and this upward sweep is followed by a downward sweep to low values. The data of Fig. 2a show that faster apping leads to faster swimming, although F(f) does not always change continuously. For A 10 cm, for example

(red curve), F increases smoothly until a critical frequency near f 1 Hz, at which F abruptly increases, in this case doubling its

speed. As f is increased further, F again increases continuously. For decreasing f, F remains high before abruptly dropping near f 0.2 Hz. Thus, the measured dynamics display a hysteresis

loop, and different apping amplitudes lead to other loops. This hysteretic behaviour is associated with bistability of states: for identical apping kinematics, the array can take on one of two gears corresponding to slow and fast swimming modes.

To organize these observations, we rst note that the quantity f/F represents the number of apping strokes in one orbit of the tank, and thus the number of strokes separating the two wings is given by S f/2F. Equivalently, this schooling number

S represents the separation distance between successive swimmers measured in wavelengths of motion, and this quantity serves as a general way to characterize arrays of synchronized swimmers. Specically, S encodes the spatial phase shift between successive swimmers: whole integer values of S denote spatially in-phase states in which neighbouring wings trace out the same path through space, and half-integer values indicate out-of-phase states. In Fig. 2b, we show how S depends on f for all the data of Fig. 2a. At each amplitude, low f leads to slow swimming or large S, which indicates weak interactions. Increasing f leads to decreasing S followed by saturation near a whole integer value, thus residing nearly in-phase for a broad range of frequencies. Further increasing f induces an abrupt downward jump in

S. Conversely, when f is decreased, the system approaches half-integer S (out-of-phase) before an abrupt upward jump.

Viewing all amplitude data sets together reveals preferred and avoided ranges of S for strongly interacting wing arrays. We quantify this observation in Fig. 2c, in which the histogram of S mod 1 reveals the distribution of spatial phase shifts. Values between 0.5 and 1 are almost entirely absent, and among the prevalent phases, distinct peaks appear at values of 0, 0.25 and 0.5. Thus, these swimmers preferentially occupy spatial phase relationships corresponding to purely in-phase, leading by a quarter cycle, and purely out-of-phase. The appearance of these particular phase relationships suggests that schooling in this system involves bodies interacting coherently through their wave-like ow elds.

Flow visualization. To shed light on the hydrodynamics of these modes, we seed the uid with particles and use high-speed video and image analysis to extract the ow elds (Supplementary Movies 2 and 3). First, we consider a single wing swimming at relatively low amplitude and frequency, in which case interactions are weak. Indeed, the ows produced are largely dissipated between successive passes, and thus the wing swims into quiescent uid, as shown by the velocity vector eld of Fig. 3a and the schematic of Fig. 3b. The wing leaves behind a stream-wise array of counter-rotating vortices that are staggered vertically. This so-called reverse von Karman wake has been observed in biological and physical experiments, and is a signature of thrust production during apping3338. The vortices are separated by strong up and down ows, which we highlight in Fig. 3a using a colour map of the vertical component of velocity. Each upstroke generates an upward ow (red) and each downstroke a downward ow (blue), and thus the wake can be viewed as a wave in space that reects the trajectory of the wing.

To induce stronger interactions, we attach multiple wings to the assembly but maintain the same apping motion. Under these conditions, the system exhibits a hysteresis loop associated with the slow and fast swimming modes. Figure 3c and d show the ow eld for the slow mode, in which successive wings trace out similar paths through space. Captured here during the downstroke, a wing is seen to sink within the downward (blue) ow generated by its predecessors. Subsequently, the upstroke occurs within an upward (red) ow. This visualization reveals that not only are successive wings spatially in-phase with one another but also any given wing is also in-phase with the existing ow structure into which it swims. Thus, this constructive wingwake interaction mode involves the repeated reinforcement of existing ows. Conversely, the fast mode

b

c

0.25

0.75

Rotational frequency, F (Hz)

a 1st Wing 2nd Wing

A =10 cm 4 cm

3 cm

2 cm

1.5 cm

Flapping frequency, f (Hz)

Schooling number, = f/2F

0 0 1 4

3 0 1 2 4

3

1

5

0

S mod 1

0

4

3

0.5

0.5

2

1

2

Flapping frequency, f (Hz)

Figure 2 | Dynamics of interacting wings. (a) Swimming speed, as measured by the rotational frequency F, versus apping frequency f. For each peak-to-peak amplitude A, an upward sweep of f is followed by a downward sweep (as indicated by arrows), and the data form a hysteresis loop. (b) Schooling number S, which represents the number of wavelengths separating successive wings. Each hysteresis loop is bounded by in-phase (integer value of S) and out-of-phase states (half-integer S). (c) A polar histogram of S mod 1 shows peaks corresponding to preferred spatial phase relationships between successive wings.

NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514

a c e

b d

f

Figure 3 | Flow visualization. (a,b) Flow eld around a non-interacting wing of chord length 3 cm extracted using particle image velocimetry and rendered in a schematic. The colour map indicates the vertical component of the velocity vector eld, with red indicating upward and blue downward ows. The upstroke produces an upward ow and the downstroke a downward ow. (c,d) Slow mode of interacting wings: the downstroke of a wing (red path) occurs within the downward ow of its predecessor (blue path). (e,f) Fast mode: the downstroke occurs within the upward ow of its predecessor.

Reynolds number, Ref

3 a

3

In-phase array Out-of-phase array

c

e

Schooling

number, S

Schooling

number, S

2

2

1

1

0

Normalized power, P

Normalized power, P

0.5

b

f

1

d

1.5

1

Isolated wing Increasing Ref

Decreasing Ref

0.5

0 20 40 80

60

0 20 40 80

60

Reynolds number, Ref

Figure 4 | Simulations of interacting wings. (ad) An innite array of synchronized or temporally in-phase wings is simulated by a single airfoil driven to ap up and down, and allowed to swim freely left to right across a periodic domain. (a) Schooling number S for increasing (blue) and decreasing (red) apping Reynolds number, Ref. A non-interacting wing (dashed curve) swims at a speed intermediate between the two schooling modes. (b) Input power normalized by that of an isolated wing. (c) Computed vorticity eld for the slow mode (blue circle in a): The wing slaloms between vortices. (d) Fast mode (red circle in a): the wing intercepts each vortex core. (e,f) Schooling dynamics and power consumption for an array in which nearest neighbours ap temporally out-of-phase with one another.

involves antagonistic motions between each wing and the ow it encounters. As shown in Fig. 3e,f, the downstroke of the wing moves against the upward ow of its predecessor, and the subsequent upstroke occurs within a downward ow. This destructive mode thus involves the inversion of ow elds, in which the passing of a wing replaces an existing upward ow with a downward one, for example.

Simulations. To gain additional insights into the intrinsic dynamics of swimmer arrays, we conduct computational uid dynamics simulations that solve for the ow eld and locomotion of a apping body. Similar to the experiments, these simulations involve free swimming of a wing, where the emergent speed reects hydrodynamic interactions (see Methods section as well as Supplementary Movies 4 and 5). The swimming dynamics is determined from the computed uid forces, and an innite array of swimmers is replicated by having a single wing repeatedly traverse a domain of length L with periodic boundary conditions39. Unlike the experiments, the ow eld is 2D and the swimming motion is translational rather than rotational. Thus, observations common to both experiments and simulations are expected to be generic and not due to system dimensionality or geometry. We systematically explore the dynamics through a procedure similar to that of the experiments: Flapping frequency is incrementally increased, with each step initialized with the nal

data of the previous step and then allowed to reach a terminal swimming speed. This upward sweep of frequency is then followed by a downward sweep.

These simulations reveal a number of features in common with the experiments, including the coexistence of different locomotion states for identical apping motions. Figure 4a shows the complete characterization of the dynamics using S f/F,

where F U/L is the traversal frequency. Here, the simulation

results depend only on apping Reynolds number Ref, which plays a role analogous to f in experiments (see Methods). Dynamical signatures of strong interactions again include bistability of states, a hysteresis loop, and avoided values of S, suggesting that these are general features of locomotor arrays. Similar to the experiments, these simulations display a downward jump at whole integer S, but unlike the experiments the upward jump seems to occur at a quarter rather than half integer, leading to a wider band of avoided phases. Nonetheless, the ow elds at these transition points reveal constructive and destructive characters, respectively. For the slow mode near RefE60 and

SE2, the vorticity plot of Fig. 4c shows that the wing slaloms between vortices, contributing vorticity of the same sign to each. For the fast mode near RefE20 and SE1.25, Fig. 4d shows that the wing intercepts and destroys each oncoming vortex core. Interestingly, previous 2D experiments of apping foils in unsteady ows have also shown vortex slaloming and interception modes40 (see Discussion section).

4 NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514 ARTICLE

a

Importantly, these simulations allow us to compare the swimming speed and power input for interacting bodies versus an isolated, non-interacting wing. The latter case is simulated in a longer domain to ensure that terminal speed is reached in the absence of self-interactions (see Methods section). To compare speed, we plot as a dashed curve in Fig. 4a the computed S for an isolated wing, which is found to lie between the curves for the slow and fast modes. Thus, with respect to speed, hydrodynamic interactions can be benecial or detrimental. To compare energetic cost, we use the computed vertical force and prescribed vertical velocity to form the average input power at steady state, and Fig. 4b shows the power investment in the presence of interactions normalized relative to that of a non-interacting wing at the same Ref. Surprisingly, both the slow and fast modes save power for nearly all Ref, and the presence of interactions can lead to savings of 450%. The slow mode always saves more than the fast, and the fast mode displays a moderate energetic increase relative to an isolated wing for a narrow range of Ref near the upward transition point.

These simulations also allow us to explore how differences in temporal phases between swimmers in an array affect their dynamics and power consumed. We consider a linear array in which each wing aps out-of-phase in time with its nearest neighbours (see Methods section for details). To compare these results to the previous case of a temporally in-phase array, we use a natural generalization of the schooling number based on the total phase shift due to both the temporal phase jT and spatial phase contributions: S jT/2p f/F. Temporally in-phase arrays

have jT 0 and out-of-phase arrays have jT p, and for

all cases this denition preserves the property that integer values of S represent spatially in-phase trajectories while half-integer S indicate spatially out-of-phase trajectories. In Fig. 4e, we characterize the locomotion of out-of-phase arrays using this modied schooling number, and we nd that all the salient features of in-phase arrays are also present in this arrangement. Interestingly, important differences between in-phase and out-of-phase arrays arise in the power consumed, as seen by comparison of Fig. 4b,f. In particular, the fast mode for out-of-phase arrays is associated with higher power consumption, and indeed it is typically higher than that of an isolated swimmer. The origin for this difference is unclear but may be related to the more erratic and intense ows observed for the case of temporally out-of-phase arrays.

Mathematical model. Our experiments and simulations motivate a minimal model that describes the collective dynamics of a linear array of swimmers. As shown in Fig. 5a, an innite array of bodies apping in synchrony and spaced by a distance L is represented by a single body that repeatedly traverses a domain of length L that is specied by periodic boundary conditions. In our conception, the bodys horizontal speed is perturbed because it encounters the wake produced in its previous pass through the domain. The perturbation strength depends on the traversal time t t0, which is the time elapsed since the body was last at

the same location: X(t) X(t0) L. Models of this type take

the form of a delay differential equation for the swimming speed _

X U U0 DUt t0. Here, U0 is the speed in the

absence of interactionsthat is, the speed of a single, isolated swimmerand DU represents the perturbation due to wingwake interactions. The effect of memory is explicitly incorporated through the time delay t0, which is not a constant but rather depends on the dynamical history.

Here we consider a specic model of this type given by the equation:

_

X sf p ee t t0=tcos 2pf t t0

: 1

The rst term describes how the speed of an isolated swimmer increases with apping frequency f, where s and p are parameters. This power law dependence of speed on frequency is consistent with our measurements for a single wing. The second term represents the perturbation to the speed, where e is the wingwake interaction strength. Importantly, the perturbation depends on the difference 2pf(t t0) in the current phase in the

apping cycle and the phase when last at the same location. One might expect that the forcing is a periodic function of this phase difference, and the cosine form, in particular, is found to yield model solutions that closely correspond to the experimental data (see below). Finally, the dissipation of ows, and thus weakening of interactions for longer traversal times, is captured by the exponential term with a decay timescale of t.

We then seek steady swimming solutions corresponding to _

X LF, where F 1/(t t0) is the frequency with which the body

crosses the domain. Putting these relationships in the above

dynamical equation and taking L 1, we obtain a nonlinear

algebraic equation relating F and f:

F sf p ee 1=tFcos 2pf =F

2nd Particle

L

L

1st Pass 2nd Pass

1st Particle

b

1.5

Schooling number, = f/F

1

0.5

Non-interacting Stable Unstable

0 0 0.5 1 1.5Frequency, f (Hz)

Figure 5 | Mathematical model. (a) An innite linear array of synchronized swimmers is represented by a single particle undergoing repeated passes across a domain specied by periodic boundary conditions. (b) Schooling number for a model with parameter values s 1.5, P 1.5, e 1, t 1 (see

text for details).

: 2

To illustrate the structure of the solutions, which are solved numerically, we display in Fig. 5b the schooling number S f/F

for a model with order-one parameter values, as given in the caption of Fig. 5. The solution curve S(f) displays a fold that consists of upper and lower stable branches (solid curves) connected by an unstable branch (dotted curve). The non-interacting swimmer (dashed curve) serves as a point of comparison.

At low f, the wing progresses slowly, S is large, and the solution resembles the non-interacting case. The fold appears at higher f and corresponds to the coexistence of a slow mode (SE1)

and a fast mode (SE0.5), and at yet higher f the slow mode disappears. The speed of an isolated wing lies intermediate between these modes, as found in simulations. The model thus reproduces remarkably well the key observations from the experiments and simulations, suggesting that relatively simple interaction laws that include memory underlie the complex hydrodynamics of swimmer arrays.

NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514

DiscussionCollectively, these ndings indicate that the intrinsic dynamical states of locomotor arrays involve repeated and coherent interactions between each apping body and the oscillating ow into which it swims. Physically, we interpret these modes as stable equilibria, that is, conditions for which thrust and drag balance to yield steady swimming and the system returns to its original speed if perturbed. Although the associated ow elds are spatially and temporally complex, experiments and simulations reveal a slow mode associated with constructive wingwake interactions and a faster, destructive mode. These observations motivate a dynamical model that incorporates a forcing that depends on the relative phase between oscillations of the body and the oncoming ow, and the strong correspondence with experiments suggests that the uid-mediated interaction laws have a tractable mathematical form.

The success and simplicity of our model also suggests that our ndings are rather generic, and indeed previous studies of apping bodies in unsteady ows show behaviour reminiscent of the interaction modes discussed here28,30,4045. Of particular relevance are the experiments of Gopalkrishnan et al.40, which determine the conditions for which a apping foil xed within an oncoming ow interacts constructively or destructively with the unsteady drag wake of an upstream cylinder. Our experiments show that analogous modes exist for self-propelled bodies, which interact through thrust wakes and whose speed is not prescribed but is dynamically selected. Additional novel aspects of our work include the identication of such modes as stable equilibria of locomotion and the discovery of coexistence of modes for identical driving conditions.

Many topics remain for future studies, including an explanation of the differences between our 3D experiments and 2D simulations in the ne structure of the hysteretic dynamics. Further, while our simulations show that temporal variations in swimming speed are typically o10%, future studies involving bodies of lower density relative to the uid will assess how stronger uctuations affect the dynamics. It will also be of interest to observe the modes assumed by arrays in which the inter-wing spacing is not xed but is dynamically selected. We also aim to better understand how the power consumed is linked to the locomotion modes, a topic not addressed by the experimental measurements and mathematical modelling presented here. Ultimately, a long-term goal of this line of research is a derivation of schooling interaction laws directly from the relevant hydrodynamics.

Because the intrinsic modes of interaction represent collective locomotion in the absence of active behavioural response, they provide a purely physical system to which highly ordered animal groups can be compared. Previous studies of schools have not included all of the measurements necessary to determine spatial phase; however, when swimming in the wake of a xed cylinder, sh adopt apping motions that are spatially in-phase with the oscillating ow46,47. In the language introduced here, this behaviour corresponds to integer values of the schooling number, S mod 1 0. For formation ight of birds, recent

measurements show that individuals are either out-of-phase (S mod 1 0.5) or one quarter phase (S mod 1 0.25) with

respect to the ow produced by an upstream neighbour20. In light of our results, we interpret these particular locomotor-ow phase relationships as suggesting that coherent uid-mediated interactions play a role in determining group structure. For formation ight or swimming of animals, perhaps the emergent structure represents a stable conguration in which not only speed but also inter-individual spacing are dynamically determined45. Future work might explore this idea by measuring the temporal apping phase, spacing and swimming speedand

thus schooling number Sin laboratory realizations of sh schools48. A narrow distribution of S across individuals would serve as evidence for coherent uid-mediated interactions.

It is also interesting to speculate how the coexistence of, and transitions between, locomotion states might impact animal collectives. Proximity to a transition point would facilitate abrupt changes in the dynamics of the entire group49, which could play a role in the rapid and seemingly coordinated response of a school to a predator, for example. On the other hand, aggregates may be susceptible to phase separation or division into slow and fast subgroups, leading to a loss of group cohesion due to fracture or collisions.

Viewing schooling as the interaction of locomoting bodies through a uid, our results highlight the critical importance of memory or history dependence. That is, the past state of the system is encoded in the ow eld, and the present state is driven according to this stored information. This effect is explicitly included in our model as the time-delayed interaction term and manifested in the hysteretic dynamics. In contrast, history dependence is not relevant to the swarming of micro-organisms and active micro-particles, where the low Reynolds number and absence of inertia ensure effectively instantaneous interactions47. This effect is also not included in current statistical mechanical models of animal collectives, which instead specify interaction rules that depend only on the current state and act instantaneously5053. Incorporating memory into such models could allow the eld of active matter to be extended to systems in which inertia is important. We also note that memory effects can be important even for the case of a single locomoting body33,36,43,54,55. For apping ight, for example, this might take the form of interactions between a wing with the vortices generated during previous strokes54.

Finally, we note that some elements of the schooling problem considered here may be shared with other systems that involve the close interaction of a body with a wave it produces. For example, intriguing parallels can be drawn to a recently studied model system that consists of uid droplets bouncing on a vertically vibrating bath24. Vibration maintains an air layer between the drop and surface, and prevents coalescence, and the bouncing motion induces waves on the bath through which droplets can inuence one another. Interaction effects include orbital motions of a pair, and memory is manifested even by a single droplet that, under appropriate conditions, can coherently interact with its own self-generated wave to induce walking or horizontal motion2426. The dynamical description of this problem shares salient features with our mathematical model, including memory-dependent forcing and coherent bodywave interactions26. In-phase schooling states in our rotational experiments are also reminiscent of the Bohr model of the atom, which associates electron orbits around the nucleus with standing wave modes. Indeed, as in the pilot-wave interpretation of quantum mechanical systems56, a self-interacting swimmer can be viewed as having particle (wing) and wave (wake) identities whose mutual inuence leads to coherent states of motion.

Methods

Experiments. Experiments use 3D-printed plastic airfoils (NACA0017) of chord length c 6 cm and span 15 cm. Wings are centred at a distance of 26 cm from the

driving axle, and the tank measures 46 cm in radius and height. Similar to bird ight and sh swimming, the wings experience high apping Reynolds numbers,Ref rfAc/mB102 104, and typical parameters include frequency fB0.1 5 Hz

and peak-to-peak amplitude AB1 10 cm. Water density and viscosity are

r 1 g cm 3 and m 10 2 g cm 1 s-1. Rotational speed around the tank is mea

sured using an optoelectronic encoder. Flow visualization is carried out in a smaller system: wings of chord c 3 cm are held at a distance 8.5 cm in a clear tank of radius

and height 15 cm. To visualize the ow elds, we seed the water with micro-particles, illuminate a cross-section of the tank with a laser sheet, capture high-speed video of the particle motion and apply a particle image velocimetry algorithm57.

6 NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514 ARTICLE

Simulations. The simulations use a Fourier spectral method with volume penalization58,59 to solve the 2D NavierStokes equations for a uid of density r and viscosity m. A NACA0010 airfoil of chord c and density 10 r is given a prescribed vertical motion of (A/2) cos 2pft, with peak-to-peak amplitude A 2c. The

integrated uid forces and Newtons second law determine the horizontal dynamics of the wing. The horizontal dimension has periodic boundary conditions, and the vertical dimension has height 3.5c and is bounded by walls. With these specications, the results depend only on the apping Reynolds numberRef rfAc/mB10 102. The ow equations are solved in the frame of the wing on

a 256 256 grid with time step 2 10 6. Simulations of a temporally in-phase

array involve a single wing traversing a domain of length L 6c. The temporally

out-of-phase case is carried out by simulating a wing pair each in a domain of length L 4c, with periodic boundary conditions being applied over the total

length of 8c. Swimming of an isolated foil is simulated using a domain of length 24c with uniform velocity enforced at the inlet and outlet. The schooling number for an isolated wing (dashed curves in Fig. 4a,e) is dened to be S jT/2p fL/U, where

jT is the temporal phase (0 for in-phase, p for out-of-phase), U is the terminal speed and L is the inter-wing spacing or domain length used in each case.

References

1. Guazzelli, E. & Hinch, J. Fluctuations and instability in sedimentation. Ann. Rev. Fluid Mech. 43, 97116 (2011).

2. Shaw, R. A. Particle-turbulence interactions in atmospheric clouds. Ann. Rev. Fluid Mech. 35, 183227 (2003).

3. Jackson, R. The Dynamics of Fluidized Particles (Cambridge Univ. Press, 2000).4. Ramaswamy, S. The mechanics and statistics of active matter. Annu. Rev. Condens. Matter Phys. 1, 323345 (2010).

5. Koch, D. L. & Subramanian, G. Collective hydrodynamics of swimming microorganisms: living uids. Ann. Rev. Fluid Mech. 43, 637659 (2011).

6. Marchetti, M. C. et al. Hydrodynamics of soft active matter. Rev. Mod. Phys. 85, 1143 (2013).

7. Saintillan, D. & Shelley, M. J. Instabilities and pattern formation in active particle suspensions: kinetic theory and continuum simulations. Phys. Rev. Lett. 100, 178103 (2008).

8. Zhang, H., Beer, A., Florin, E.-L. & Swinney, H. L. Collective motion and density uctuations in bacterial colonies. Proc. Natl Acad. Sci. USA 107, 13626 (2010).

9. Masoud, H. & Shelley, M. J. Collective surng of chemically active particles. Phys. Rev. Lett. 112, 128304 (2014).

10. Brady, J. F. & Bossis, G. Stokesian dynamics. Ann. Rev. Fluid Mech. 20, 111157 (1988).

11. Parrish, J. K. & Edelstein-Keshet, L. Complexity, pattern, and evolutionary trade-offs in animal aggregation. Science 284, 99101 (1999).

12. Sumpter, D. J. T. The principles of collective animal behaviour. Philos. Trans. R. Soc. Lond. B Biol. Sci. 361, 522 (2006).

13. Partridge, B. L. The structure and function of sh schools. Sci. Am. 246, 114123 (1982).

14. Bajec, I. L. & Heppner, F. H. Organized ight in birds. Anim. Behav. 78, 777789 (2009).

15. Weihs, D. Hydromechanics of sh schooling. Nature 241, 290291 (1973).16. Lissaman, P. B. S. & Shollenberger, C. A. Formation ight of birds. Science 168, 10031005 (1970).

17. Higdon, J. J. L. & Corrsin, S. Induced drag of a bird ock. Am. Nat. 112, 727744 (1978).

18. Hummel, D. Aerodynamic aspects of formation ight in birds. J. Theor. Biol. 104, 321347 (1983).

19. Weimerskirch, H., Martin, J., Clerquin, Y., Alexandre, P. & Jiraskova, S. Energy saving in ight formation. Nature 413, 697698 (2001).

20. Portugal, S. J. et al. Upwash exploitation and downwash avoidance by ap phasing in ibis formation ight. Nature 505, 399402 (2014).

21. Weihs, D. in Swimming and Flying in Nature, 703718 (Springer, 1975).22. Svendsen, J. C., Skov, J., Bildsoe, M. & Steffensen, J. F. Intra-school positional preference and reduced tail beat frequency in trailing positions in schooling roach under experimental conditions. J. Fish Biol. 62, 834846 (2003).

23. Partridge, B. L. & Pitcher, T. J. Evidence against a hydrodynamic function for sh schools. Nature 279, 418419 (1979).

24. Couder, Y., Protire, S., Fort, E. & Boudaoud, A. Dynamical phenomena: walking and orbiting droplets. Nature 437, 208 (2005).

25. Protiere, S., Couder, Y., Fort, E. & Boudaoud, A. The self-organization of capillary wave sources. J. Phys. Condens. Matter 17, S3529 (2005).

26. Oza, A. U., Rosales, R. R. & Bush, J. W. M. A trajectory equation for walking droplets: hydrodynamic pilot-wave theory. J. Fluid Mech. 737, 552570 (2013).

27. Wu, T. Y. & Chwang, A. T. in Swimming and Flying in Nature, 687702 (Springer, 1975).

28. Streitlien, K., Triantafyllou, G. S. & Triantafyllou, M. S. Efcient foil propulsion through vortex control. AIAA J. 34, 23152319 (1996).

29. Rockwell, D. Vortex-body interactions. Ann. Rev. Fluid Mech. 30, 199229 (1998).

30. So, R. M. C., Jadic, I. & Mignolet, M. P. Fluidstructure resonance produced by oncoming alternating vortices. J. Fluids Struct. 13, 519548 (1999).

31. Allen, J. J. & Smits, A. J. Energy harvesting eel. J. Fluids Struct. 15, 629640 (2001).

32. Whittlesey, R. W., Liska, S. & Dabiri, J. O. Fish schooling as a basis for vertical axis wind turbine farm design. Bioinspir. Biomim. 5, 035005 (2010).

33. Vandenberghe, N., Zhang, J. & Childress, S. Symmetry breaking leads to forward apping ight. J. Fluid Mech. 506, 147155 (2004).

34. Vandenberghe, N., Childress, S. & Zhang, J. On unidirectional ight of a free apping wing. Phys. Fluids 18, 014102 (2006).

35. Spagnolie, S. E., Moret, L., Shelley, M. J. & Zhang, J. Surprising behaviors in apping locomotion with passive pitching. Phys. Fluids 22, 041903 (2010).

36. Alben, S. & Shelley, M. Coherent locomotion as an attracting state for a free apping body. Proc. Natl Acad. Sci. USA 102, 1116311166 (2005).

37. Blickhan, R., Krick, C., Zehren, D., Nachtigall, W. & Breithaupt, T. Generation of a vortex chain in the wake of a subundulatory swimmer. Naturwissenschaften. 79, 220221 (1992).

38. Triantafyllou, M. S., Triantafyllou, G. S. & Yue, D. K. P. Hydrodynamics of shlike swimming. Ann. Rev. Fluid Mech. 32, 3353 (2000).

39. Hemelrijk, C. K., Reid, D. A. P., Hildenbrandt, H. & Padding, J. T. The increased efciency of sh swimming in a school. Fish Fish. 16, 511521 (2015).

40. Gopalkrishnan, R., Triantafyllou, M. S., Triantafyllou, G. S. & Barrett, D. Active vorticity control in a shear ow using a apping foil. J. Fluid Mech. 274, 121 (1994).

41. Beal, D. N., Hover, F. S., Triantafyllou, M. S., Liao, J. C. & Lauder, G. V. Passive propulsion in vortex wakes. J. Fluid Mech. 549, 385402 (2006).

42. Akhtar, R. Mittal, Lauder, G. V. & Drucker, E. Hydrodynamics of a biologically inspired tandem apping foil conguration. Theor. Comp. Fluid Dyn. 21, 155170 (2007).

43. Wang, Z. J. & Russell, D. Effect of forewing and hindwing interactions on aerodynamic forces and power in hovering dragony ight. Phys. Rev. Lett. 99, 148101 (2007).

44. Ristroph, L. & Zhang, J. Anomalous hydrodynamic drafting of interacting apping ags. Phys. Rev. Lett. 101, 194502 (2008).

45. Zhu, X., He, G. & Zhang, X. Flow-mediated interactions between two self-propelled apping laments in tandem conguration. Phys. Rev. Lett. 113, 238105 (2014).

46. Liao, J. C., Beal, D. N., Lauder, G. V. & Triantafyllou, M. S. Fish exploiting vortices decrease muscle activity. Science 302, 15661569 (2003).

47. Liao, J. C. A review of sh swimming mechanics and behaviour in altered ows. Philos. Trans. R. Soc. Lond. B Biol. Sci. 362, 19731993 (2007).

48. Katz, Y., Tunstrm, K., Ioannou, C. C., Huepe, C. & Couzin, I. D. Inferring the structure and dynamics of interactions in schooling sh. Proc. Natl Acad. Sci. USA 108, 18720 (2011).

49. Becco, C., Vandewalle, N., Delcourt, J. & Poncin, P. Experimental evidences of a structural and dynamical transition in sh school. Physica A 367, 487493 (2006).

50. Breder, C. M. Equations descriptive of sh schools and other animal aggregations. Ecology 35, 361370 (1954).

51. Reynolds, C. W. in Proceedings of the 14th annual conference on Computer graphics and interactive techniques (ACM SIGGRAPH Comp. Graph) 2534 (New York, USA, 1987).

52. Vicsek, T., Czirk, A., Ben-Jacob, E., Cohen, I. & Shochet, O. Novel type of phase transition in a system of self-driven particles. Phys. Rev. Lett. 75, 1226 (1995).

53. Lopez, U., Gautrais, J., Couzin, I. D. & Theraulaz, G. From behavioural analyses to models of collective motion in sh schools. Interface Focus 2, 693707 (2012).

54. Dickinson, M. H., Lehmann, F.-O. & Sane, S. P. Wing rotation and the aerodynamic basis of insect ight. Science 284, 19541960 (1999).

55. Li, C., Umbanhowar, P. B., Komsuoglu, H., Koditschek, D. E. & Goldman, D. I. Sensitive dependence of the motion of a legged robot on granular media. Proc. Natl Acad. Sci. USA 106, 30293034 (2009).

56. Bush, J. W. M. Quantum mechanics writ large. Proc. Natl Acad. Sci. USA 107, 17455 (2010).

57. Stamhuis, E. J. Basics and principles of particle image velocimetry (PIV) for mapping biogenic and biologically relevant ows. Aquatic Ecol. 40, 463479 (2006).

58. Mittal, R. & Iaccarino, G. Immersed boundary methods. Annu. Rev. Fluid Mech. 37, 239261 (2005).

59. Kolomenskiy, D. & Schneider, K. A Fourier spectral method for the Navier-Stokes equations with volume penalization for moving solid obstacles.J. Comp. Phys. 228, 56875709 (2009).

Acknowledgements

We thank S. Childress, F. Fang, J.M. Huang, A. Oza, S. Ramananarivo and J. Zhang for their useful discussions. MS and HM acknowledge support from United States Department of Energy (DOE) grant DE-FG02-88ER2505, and HM acknowledges National Science Foundation (NSF) grant DMR-0844115 and the Institute for Complex Adaptive Matter.

NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9514

Author contributions

L.R. and M.S. designed the study; A.D.B., H.M., J.W.N. and L.R. performed the experiments; H.M. and M.S. carried out the simulations; L.R. and M.S. formulatedthe mathematical model; and all authors contributed to analysing the data, interpreting the results and writing the paper.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Becker, A. D. et al. Hydrodynamic schooling of apping swimmers. Nat. Commun. 6:8514 doi: 10.1038/ncomms9514 (2015).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

8 NATURE COMMUNICATIONS | 6:8514 | DOI: 10.1038/ncomms9514 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.