arXiv:0908.2769v1 [physics.bio-ph] 19 Aug 2009 flwapiuewvso h ufc fterbodies their of surface the propagation on waves the amplitude to low instead of resort bacteria unflagellated imt wma 5daeesprsecond per diameters 35 at swim to nism tt wmtruhtegaua flow granular the through the swim allow is and to sand undu- example it surrounding whose the One scales), fluidize effectively smooth lations with gravity. forms lizards underground, of (small even of sandfish effects or variety gravitational ground the a the with to minimizing adopted deal close have them thus keep animals that and the Other support animal Larger that forces. the limbs system. of to to motor resort weight small may their too animals in are terrestrial components the life Microorganisms many and non-plant forag- have most locomotion, diverse. for of danger, are Whether mechanisms avoid solutions evolved to have movement. or forms is mating for life ing, animal and bial a ael n a,wietecl oyrttsi h op- the in rotates body propeller. cell a direction. heli- the posite their while as way, all that one act rotating flagella by and flagella, cal swim membrane called bacteria flagellated cell filaments, These the helical from oars long use little extend more bacteria like Motile or much water. used one through are Paramecia, row that synchronously (protozoa), to cilia undulate. organisms with eukaryotic to covered celled are shape, single body many microorganisms membrane, like the celled cell the modulate single from extend Even or that filaments body. elastic use a only limbs, that remarkably locomotion is body, funda- undulatory the of along widespread. these, number propagation wave small Of on a relies into classes. divided locomotion mental of be dif- all very probably yet, with can And contend environments. to physical have ferent species flying and rainean .TEUIESLT FUNDULATORY OF UNIVERSALITY THE 1. nteaia igo,lmsmgtcm nhandy in come might limbs kingdom, animal the In n ftecuileitnilrqieet fmicro- of requirements existential crucial the of One owne.T nuae ytmde o require not does system a undulate, To wonder. No plctosrnigfo oi at ipslt minimall m to that disposal undulators waste toxic microrobotic gen from in are ranging undulations b progress applications how control, recent understand motor review to system, we i considered nervous locomotion be the of all cases which must simple in example some animals, a selected to concepts small to discussion key applied the some results extend review we and We regime numbers tools tractable. Reynolds simulation more low as much the under be in the can undulators generate ment on to even large cases, and some by s in the and which environment, propulsion, physical ate the with interplay body’s oiebcei oggni rhsoi nks e ounde a to primitive ac Key is prevalent it snakes. so prehistoric locomotion is gigantic it of to that mode bacteria then, a motile wonder As No body. robust. remarkably a along waves of nuaoylcmto samaso efpouso htre that self-propulsion of means a is locomotion Undulatory .coli E. LOCOMOTION colo optn,Uiest fLesL29T ed,Uni Leeds, 9JT, LS2 Leeds of University Computing, of School o xml sssc mecha- a such uses example for , colo optn n nttt fSsesadMmrn Bi Membrane and Systems of Institute and Computing of School nvriyo ed,LesL29T ntdKingdom United 9JT, LS2 Leeds Leeds, of University 1 aie subter- Marine, . 2,3 nuaoyLocomotion Undulatory oemotile Some . Dtd 8Ags 2009) August 18 (Dated: odnH Boyle H. Jordan et Cohen Netta 4,5 . ain eyo togrrssac osdwy motion. sideways to resistance stronger undu- on (retrodgrade) rely conventional moved lations is contrast, forwards it In when moved generating effect is sideways. long minimal oars, having body the but like the backwards, to much or when angles forces act right drag protrusions significant at These out jut axis. which ‘bristles’ with Ochromonas flcmto a eahee,oems rtntc key of a notice body first the must between one achieved, similarity be mode can counter-intuitive locomotion a such of how understand To gation. eea rtza uha h oiealga motile the as malhamensis such protozoa, several raimmvsi the in moves organism ocs u ilg sfl fecpin.Sm annelids, Some exceptions. worm of polychaete full the body is including the Biology forward if But applies words, environment forces. other the tail backward, In to given travels head wave motion). from a forward (i.e., achieve in motion of propagate to move direction that the to required, to Typically, opposite are waves mammals retrograde marine down. other direction, and and from up whales bodies whereas undulate their side, flap longi- or to usually and motion) side Fish motion) hori- of transverse. of or into direction direction tudinal the classified the (in be (opposite direct retrograde crudely vertical, can or These zontal locomote. to htntol wmbtaeas aal fburrowing of capable also mud are or a but eels, through sand swim move as through only (such to environments not them physical allow that different to And of gaits adaptability variety like nooks. and undulations. flexibility crawling narrow the the into of exploit complement im- burrow may to species against an to many them them need of use may press more Crawlers or to are body choose they the can when swimmers but pediment, purposes, some for ose 1mln)peitrcttnbasnakes the titanoboa to prehistoric microbes tiny long) even from (13m and size monster fish, in in snakes, ranging locomotion lizards, mammals of worms, some larvae, means of primary variety the a as undulations find nml s ait fdffrn om fundulation of forms different of variety a use Animals ∗ † naiesurgery. invasive y imro rwe ilepott gener- to exploit will crawler or wimmer fbooia wmes nparticular, In swimmers. biological of s sadn nuaoylcmto sthe is locomotion undulatory rstanding yn nuain.Ti eiwfocuses review This undulations. lying iso h eeainadpropagation and generation the on lies o-etna ei swl sto as well as media non-Newtonian n yoedybcm omnlc in commonplace become day one ay osarneo ilgclsae from scales biological of range a ross d rprisadteenvironment the and properties ody rtdadmdltd oconclude, To modulated. and erated hr h hsc fteenviron- the of physics the where , nbt ae,tebd/aelmi lined is body/flagellum the cases, both In . 9 drltvl ipe e a be can yet simple, relatively nd dtertclavne,a well as advances, theoretical nd xii h eakbepoet htthe that property remarkable the exhibit e Kingdom ted 6 same .Tu,i sntsrrsn to surprising not is it Thus, ). ology, ieto stewv propa- wave the as direction Nereis eesvirens Nereis n h ael of flagella the and Ochromonas 8 swl as well as , 7 . 2.1 What is the Reynolds number? 2 THEORY

If there is a unifying principle to all of the above, it is dragging and indicates how nearby slices of fluid are dif- that undulating motion is typically constrained by fric- ferentially dragged95. In fact, not all fluids obey the lin- tional or drag forces of the environment, rather than by earity of Eq. (1), but those that do are called Newtonian gravitational forces. For instance, most swimmers can fluids. Examples include air, water and indeed honey float in water using buoyancy, but self-propulsion be- with respective viscosities of O(10−5)Pa·s, O(10−3)Pa·s comes much more energy intensive than in air or on the and O(104)Pa·s. ground. This is all the more severe in turbulent waters Now if we change the speed with which we are moving or when swimming upstream in a river. Fish use a range the spoon (or stop it altogether), we must also consider of individual and group strategies to exploit the hydro- the inertia of the fluid. Inertial effects will dominate over dynamics and minimize energy expenditure. Many fish, viscous forces when the spoon is sufficiently fast, or alter- for example, can minimize power expenditure (i.e., mus- natively, when the fluid has sufficiently low viscosity. In cle work) by recycling energy from vortices in turbulent this case, the force applied by the spoon on the fluid can flows. In fact, by swimming in schools, trout kinematics result in nonlinear convective and turbulent flows, which are not so different from that of passive hydrofoils (mim- 10,11,12 will then feed back and influence the motion. Let us try icking flag waving motion) . Terrestrial crawlers to determine when such inertial effects are important. (and swimmers1,13) keep close to the ground, so they are To do so, consider the motion of an object through a much more constrained by the terrain and its associated Newtonian fluid. Suppose the velocity v of the fluid drops frictional forces than by gravity. But this is perhaps most off linearly away from the object. The viscous forces in obvious in the so called low Reynolds number regime, the fluid around the object are given by µAdv/dy, which where inertial forces become altogether negligible. In all will then scale as O(µℓ2v/ℓ)= O(µℓv), where ℓ is a char- of the above, the immediate implication is that undula- acteristic size of the object. The inertial forces (due to tory locomotion arises from the interaction between the the fluid’s momentum in the same region mdv/dt) should dynamics of the body and the physics of the environment scale as O(ρℓ2v2) where ρ is the density of the fluid. The and hence places strong constraints on the shape of the ratio of these two expressions is then characterized by a body. single dimensionless scaling parameter

ρℓ2v2 ρℓv Re = = , (2) 2. THEORY µℓv µ

In this paper, we will try to restrict ourselves to some where Re is the conventional shorthand for the ‘Reynolds of the simplest cases of undulatory locomotion operating number’. When Re ≫ 1 inertial forces dominate. By in the low Reynolds number regime and where two di- contrast, if Re / 1 the viscous forces dominate the flow mensional treatments suffice; body shapes are long and and the fluid largely responds to external forces in a pas- sive manner. To give ballpark figures, a person swimming slender, and the environment is relatively simple. We 4 begin with an introduction to some of the theoretical in water might experience a Reynolds number of O(10 ). If we tried to swim through honey, we might feel a Re foundations in fluid mechanics and their applications to −3 self-propulsion in general and to undulatory locomotion around O(10 ) and bacteria swimming in water may in particular. feel Reynolds numbers as low as O(10−5)! The Reynolds number can also be obtained from the governing equation in fluid dynamics, the so-called Navier-Stokes equation (given here in simpler form for 2.1. What is the Reynolds number? incompressible fluids, i.e., ∇ · v = 0)

∂v Fluid mechanics has long been of interest to physi- − ∇p + µ∇2v = ρ + ρ(v · ∇)v . (3) cists. Already Isaac Newton postulated how fluids of ∂t different consistencies respond to forces. Perhaps, when Isaac Newton took a break from pondering the motion of Here, the left hand side describes pressure and viscous falling apples, he was holding a spoon over his cup of af- terms, and the right hand side describes inertial terms, ternoon tea, and dragging it along the surface. He would which vanish at low Re96. In fact, it is easy to see here, have noticed that the top layer of the fluid was dragged that for an object with a characteristic length ℓ, we can along. How much more difficult would this be if he had recover the Reynolds number as the ratio of the inertial done the same across a jug of honey? Newton postulated term ρ(v · ∇)v to the viscous drag term µ∇2v. In most that the force required to keep a flat spoon moving at of what follows, we will need only the low Re reduction constant speed v would follow of the Navier-Stokes equation

dv µ∇2v = ∇p . (4) F ∝ A (1) dy moving spoon

2.2. Self-propulsion in low Re environments and where A is the contact area of the spoon, and v(y) is the speed profile of different slices of fluid as one moves the scallop theorem a distance y away from the spoon. The proportionality constant is called the viscosity of the fluid (or sometimes Most of our intuition comes from our day to day ex- the dynamic or Newtonian viscosity) µ. The gradient periences of the high Re world in which inertia must be of the velocity profile dv/dy reflects the strength of the overcome. When you start off in your car, a torque is

2 ContemporaryPhysics,toappear 2.2 Self-propulsion in low Re environments and the scallop theorem 2 THEORY applied to the wheels, which in turn apply a backwards and Wilczek16 we could construct an alphabet of all pos- directed force to the surface of the road. The reactive sible shapes {S0} of the swimmer, all at the same location (forwards directed) frictional force on the car gives rise and orientation (the origin). Then the motion of a de- to acceleration. At some speed this propulsive force is formable body undergoing an ordered sequence of body- counter-balanced by wind resistance and the velocity set- shape changes S0(t) will be described by a rotation and tles. The elimination of inertia at low Re means that displacement of the body to its appropriate orientation any nonzero resultant force acting on an object will give and location S(t) via rise to infinite acceleration. Thus, somewhat counterintu- itively, the total net force and torque acting on an object S(t)= R(t)S0(t) , moving in a low Re environment will at all times be zero. As an example, suppose a small swimmer is moving at where R combines both rotation and displacement oper- constant velocity, and then stops swimming. The above ations. Now to follow shape changes in continuous time condition will result in immediate deceleration and the we would like to follow infinitessimal shape changes, and swimmer will stop. Swimming is indeed hard work at hence it is convenient to introduce an exponential form low Reynolds numbers. Curiously, low Reynolds num- (so infinitessimal generators will form a Lie algebra). In ber physics is remarkably reminiscent of the Aristotelian particular, defining view of physics, in which objects will remain stationary in the absence of external forces. Aristotle’s mechanics dR dR has long been dismissed as fundamentally flawed and su- = R R−1 ≡ RA dt dt perceded by Newtonian mechanics, so it is reassuring to   see that this theory too has found its natural place. Let us now consider a small swimmer (and hence at allows us to write low Re). To have any chance of moving, it must be able t to change its shape. The sum of all internal forces must ¯ ′ ′ clearly be zero (for the same reason you cannot lift your- R(t)= P exp A(t )dt , (5) self up by your boot straps). The change of shape will Z0  result in some motion of parts of the body in a global co- where P¯ represents ordering terms in the expansion of ordinate frame, which will then elicit reactive drag forces. 97 But since these reactive forces must sum to zero, the or- the exponent with later terms to the right . Although ganism as a whole will move in such a way that this is the A(t) is in general time dependent, the linearity of the case. This condition is sufficient to uniquely determine Navier-Stokes equation at low Re means that speed does the motion of the whole organism given a known time not matter. Indeed, as we have already argued, in the series of body configurations and known environmental absence of inertial effects, the sequence of shape changes properties. completely specified the rotation and displacement of a body. Thus, A(t) may be recast in an effectively time- In fact the low Re physics imposes constraints on the 16 possible shape changes that will result in progress. This independent geometric manner . Define a vector over was realized by Ludwig14 and then by Purcell15 who shape space, also denoted A, whose projection onto the nicely formulated it as the scallop theorem. Consider a “direction” (in shape space) dS0/dt is just A(t), via scallop that opens and closes its shell in water to move. At sufficiently high Re, the slow opening and rapid shut- A(t) ≡ AS˙0 [S0(t)] . ting of the shell pushes water out and propels the scallop in the opposite direction. At low Reynolds number, the Thus, at low Re, an integral over time can be recast as flow of water into and out of the scallop over one cy- an integral over shapes and Eq. (5) becomes cle would be the same, regardless of the speed. Scallops would make no net progress at low Re. S0(t) The derivation of this theorem is straightforward and R(t)= P¯ exp A[S0] · dS0 . instructive. We begin with the Navier-Stokes equation at S0(0) ! low Reynolds number [Eq. (4)]. Note that this equation Z is time independent. This means that speed makes no For a cyclic stroke, the net rotation and displacement are difference to the motion. Only the sequence of configu- then rations of the body determines the motion. But if that sequence is time reversible, such that moving forward or backward in time involves the same sequence of shapes, R(t)= P¯ exp A[S0] · dS0 . then no overall progress can result. I  How then is propulsion achieved at low Re? It is rea- sonable to assume that successful propulsion always re- Thus, the net progress our swimmer makes in each cy- lies on a repetitive “stride” or cyclical motion, but can we cle is proportional to the area circumscribed in shape move beyond the no-go scallop theorem towards a unify- space. For a simple back-and-forth motion such as the ing theory of shape changes that sustain self-propulsion scallop’s, no area is “cut out” and so no overall progress at low Re? can be made. This geometric formulation therefore gen- eralizes Purcell’s scallop theorem by relating the progress of a low Re swimmer to a geometric phase (akin to a classical Berry’s phase). Of course, to determine how 2.2.1. A generalized scallop theorem much progress is actually made still requires solving the fluid dynamic equations to obtain the shape space vec- Consider some cyclic motion of a low Reynolds num- tor A[S0]. This generalized formulation of the scallop ber swimmer in a Newtonian fluid. Following Shapere theorem is due to Shapere and Wilczek16.

3 ContemporaryPhysics,toappear 2.2 Self-propulsion in low Re environments and the scallop theorem 2 THEORY

2.2.2. Spherical symmetry breaking for propulsion sic equations for relatively simple but limited classes of problems. We begin with the condition for incompress- While violating the time reversibility condition of the ible fluids ∇ · v = 0. Taking the curl of Eq. (4) yields scallop theorem is necessary for successful propulsion at 2 low Re, there is one other asymmetry that needs mention- ∇ (∇× v)=0 . (6) ing. Regardless of scale or Reynolds number, successful To solve the equation for swimmers in a fluid, we need propulsion requires an asymmetry or anisotropy in the only add boundary conditions, satisfied by the fluid at environmental resistance to the motion of the body. Con- the boundary of the swimmer’s body. This is typically sider a compact shape that undergoes a time-asymmetric the no-slip condition cycle of shape changes, but remains at all times spheri- cally symmetrical. It is trivial to see that such a shape δS would go nowhere. Thus, some form of spherical symme- v|S = , (7) try breaking is needed to achieve locomotion. δt Indeed, nearly all life forms, from bacteria to mammals which requires that the fluid is perfectly dragged along have a distinct body axis or polarity, which dictates the at the boundary S. direction of motion. When an elongated body is mov- We now consider two simplifications of the Navier- ing forwards (parallel to its long axis) at low Re, it will Stokes equation for an incompressible fluid at low Re displace less fluid per unit time than when moving side- [Eq. (4)]. First, if the velocity profile v takes the form of ways (normal to its long axis). Thus, it will encounter a so-called potential flow v = ∇φ, with ∇2φ = 0, then a smaller resistance from the fluid. The ability of a de- Eq. (6) is automatically satisfied. In this case, solving the formable body to modulate the level of resistance it en- Laplace equation for a scalar field presents a vast simpli- counters in different directions allows, in principle, for fication over the Navier-Stokes equation. For example, motion to be possible. This point will be revisited in this approach allows us to prove Stokes’ Law, namely, Sec. 2.3.1. that the speed v at which a sphere of radius r will tend To demonstrate the minimal conditions for swimming, to be towed in a Newtonian fluid with viscosity µ under we give two very simple examples: one of a generalized the application of a force F is given, in the limit of small scallop that can rotate, but cannot swim (or propel itself Re, by in one direction), and another of a minimal swimmer. Consider a cyclical version of Purcell’s scallop, where F =6πµrv . (8) instead of opening and shutting, the scallop’s opening angle keeps increasing (at 2π per cycle). Shapere and Importantly, this approach can also be used to derive Wilczek16 already noted that the Scallop theorem as- the theory of slender bodies, which underlies most of our sumes that the scallop cannot turn through more than understanding of undulatory physics at low Re. 2π on its hinge). However, if we allow the scallop to do A more general simplification occurs in 2D. Then we so, and assuming the two arms are even, the scallop will may always write v = (∂U/∂y, −∂U/∂x) where U is a still get nowhere. To break time reversal symmetry, one scalar potential. Hence Eq. (6) reduces to the biharmonic arm could be longer than the other, and hence subject equation to a greater resistive drag force, as shown in Fig. 1A. In 4 this case, the scallop can rotate, but will not achieve any ∇ U =0 . (9) lasting translation (Fig. 1B). This, in addition to the no-slip condition, gives a full Consider, by contrast, the following push-me-pull-you 17 description of low Re (incompressible) fluid dynamics in example due to Avron et al. in Fig. 2. Here we have two 2D. perfectly spherical shapes that are experiencing isotropic drag at all times. However, by allowing the spheres to in- flate and deflate, the relative drag forces they experience change in time, allowing the motion of the spheres to be 2.3.1. Slender body theory asymmetrical as well. In either case, applying Shapere’s and Wilczek’s formalism above should demonstrate that Undulations are particularly appealing to study, not an area is indeed carved out in shape space, leading, in only because of their ubiquity but also because the mo- the asymmetric scallop case, to an overall rotation and, tion can be elegantly formulated as the propagation of a in the push-me-pull-you case, to an overall displacement. wave. In particular, in low Re it turns out that some (Note that the push-me-pull-you example involves two very general statements can be made within what is degrees of freedom, but some restricted forms of propul- often dubbed slender body theory. We owe our un- sion can also be achieved with only a single degree of derstanding of the physics of undulatory locomotion in freedom.) large part to the pioneering works of physicists and applied mathematicians (such as G. I. Taylor, M. J. Lighthill and G. J. Hancock) as well as zoologists (no- tably J. Gray, H. W. Lissmann and H. R. Wallace) in 2.3. Incompressible fluid dynamics at low Re the 1950s18,19,20,21,22,23,24. These authors described the locomotion, analyzed the physical forces and derived the To solve the equations of motion of a body moving mathematical framework that we still use today to under- through a fluid is a daunting task which, in general, re- stand low Re undulatory locomotion. (Incidentally, the quires a solution to the Navier-Stokes equations. When theory developed for high Re undulations bears many convective and turbulent forces dominate, this is indeed similarities to slender body theory and is called elon- arduous , but even at low Reynolds numbers the problem gated body theory25). In this section, we present a brief is rarely solved analytically. Below, we introduce the ba- overview of key results for slender body locomotion.

4 ContemporaryPhysics,toappear 2.3 Incompressible fluid dynamics at low Re 2 THEORY

A B

θ

FIG. 1: A. A variation on Purcell’s scallop in which the two arms have different lengths and are allowed to rotate freely around (like a wheel). Arms are modeled as slender cylindrical bodies. The medium is modeled as Newtonian (K = 2). The path of the scallop’s “center of mass” (CoM, red dot) is shown in black. After a few rotations, the CoM appears to undergo both translation and rotation. However, the CoM never leaves a circular area of small radius, as can be seen by following the trajectory over sufficiently many cycles (B). Each ‘petal’ corresponds to one cycle of rotation through 2π.

[analogous to Eq. (8)] as

Fi = −ci vi , (10)

where ci are the effective drag coefficients for motion tan- gent (ck) and normal (c⊥) to the local body surface. Within this framework, R. G. Cox was able to de- rive equations approximating ck and c⊥ as functions of FIG. 2: Schematic of a self-propelled push-me-pull-you de- the length and radius of the body, and the viscosity of vice, adapted from Ref. [17]. To move, the device follows a the Newtonian fluid26. Approximating the body shape cyclical sequence of steps that include: (i) shortening the con- as a prolate ellipsoid, J. Lighthill obtained similar but necting rod between the spheres; (ii) inflating and deflating slightly more accurate expressions for the effective drag the two spheres so their volumes are switched; (iii) length- coefficients27: ening the rod to its original length; and (iv) deflating and 4πµ inflating the two spheres to their original sizes. For the initial c = L conditions depicted here, this sequence will lead to motion to ⊥ ln(2q/r)+0.5 the left. 2πµ c = L , (11) k ln(2q/r)

where L is the body length, r is the body radius, q = As we introduced above, two key asymmetries are re- 0.09λ (with λ being the wavelength of the undulation) quired for an organism to be capable of low Re swimming. and µ is the viscosity of the fluid. The requirement of Not only must the undulation be asymmetric under time asymmetric drag forces can be neatly expressed in terms reversal, but some asymmetry in the environmental re- of the ratio K = c⊥/ck which must have a value other sistance is also required (the latter being a more general than unity if progress is to be possible. Notice that the requirement of locomotion at any Re). Organisms that viscosity of the fluid has no effect on this ratio. Rather, it use undulatory locomotion are generally long and thin; is completely determined by the geometry of the object. in fluids, this guarantees asymmetry in the resistance to For worm-like shapes, K typically takes values around forward (or backward) and sideways motion. 1.5, whereas for infinitely long cylinders, K approaches Now as already noted, analytically solving the motion 2 (K → 2). of non-spherical shapes in a fluid is non-trivial. Slender body theory approaches this by deriving approximate so- lutions of the Navier-Stokes equation for no-slip bound- ary conditions applied to long cylindrical or similarly 2.3.2. Undulations in rigid channels elongated shapes. These solutions take the form of force- velocity relations. Decomposing those into their vec- To understand how propulsion is generated, we con- tor components then leads to two different linear force- sider the simple case of a cylindrical organism whose velocity relations along the major and minor axis of the body undulates sinusoidally in a plane. From the scallop object. It then becomes possible to write drag equations theorem we know that a standing wave will be unable

5 ContemporaryPhysics,toappear 2.3 Incompressible fluid dynamics at low Re 2 THEORY to propel the organism, so instead consider a traveling have K → ∞, we can see that for any B, we will have wave that is propagated backward from head to tail, and vslip → 0, so that over a single period of undulation the is therefore not time symmetric. In the absence of any body travels a distance of one wavelength. Interestingly, fluid or walls (i.e., in vacuum), the body wave could still slender body theory is valid for any K > 0. For K = 1we propagate backwards at velocity vwave, but the organ- obtain 100% slip (equivalent to a vacuum, or completely ism would remain stationary. By contrast, now consider isotropic environmental resistance) and for K < 1 we ob- the limit K → ∞, which could be achieved by placing serve direct wave propagation, with the body progressing the organism in a tightly fitting sinusoidal channel with in the same direction as the wave (Fig. 3B). the same amplitude and wavelength as the body wave While approximate, the simplicity and tractability of but with rigid walls21. Within this channel, the wave is the force resistivity theory has led to its extensive appli- by definition stationary in global coordinates, forcing the cation in biological domains, in particular for the study organism forwards at a velocity vprog = −vwave. of flagellar propulsion in viscous fluids and nematode lo- Let us examine how propulsion is achieved down the comotion in viscous and visco-elastic fluids (see below). channel. As the wave is propagated, the channel will ap- In both cases, comparisons either against slender body ply a reactive force sufficient to prevent any motion in the theory30 or against data31 have concluded that the ap- normal direction. These reaction forces will only occur proximations are reasonable. on the leading (i.e., backwards facing) edge of the wave. Now, the forwards directed components of these forces will add up, yielding a net propulsive force down the V V t=0 channel while any sideways directed components will can- prog wave t=1 cel out. Hence, the organism will move forwards through the channel, with some velocity vprog. As the organism slides forwards, it will rub against the sides of the chan- K > 1 nel and evoke reactive drag forces. Again the sideways directed components will cancel out over a cycle, but the backwards directed components will sum, yielding a net V wave t=0 retarding force opposite to the direction of motion. The t=1 propulsive force exerted by the walls of the channel will be exactly sufficient to counteract this retarding force V when the organism progresses at velocity vprog = −vwave. prog K < 1

2.3.3. Low Re undulations: “slip” formulation FIG. 3: The effect of K on propulsion. An undulating body Clearly, the K → ∞ case can be trivially solved with- propagates a sinusoidal wave to the right at velocity vwave out recourse to fluid dynamics or slender-body theory. (red arrow). If K = 1, no motion will result (not shown). If Consider the same organism in a fluid, i.e., with finite K > 1, the body moves at velocity vprog in the direction op- K (the same reasoning will apply for any K > 1). posite to vwave , with |vprog| ≤ |vwave |. If K < 1, the situation Now, the normal component of velocity v⊥ must be reverses and the body moves in the same direction as vwave , non-zero if the normal force F⊥ is to be non-zero. So, with vprog < vwave . rather than the wave remaining stationary in the global frame, it will slip backwards at some velocity vslip (with |vslip| ≤ |vwave|) while the organism moves forwards at velocity vprog = −(vwave − vslip) . This is schematically illustrated in Fig. 3A. An approximation of slender body theory can be obtained from the so called force resistivity 2.3.4. Beyond Newtonian fluids theory and is due to Gray and Hancock28. Here, rather than solving the Navier-Stokes equations for a long and It is remarkable how adept biological organisms are at slender body, the forces are approximated independently adapting to different environments and modulating their at each point (using equations of the form Fi = ci vi as behavior. Many organisms exhibit enormous flexibility before) and then integrated over the body length. Thus, in navigating a wide range of environments, whether this any correlations in the fluid due to the spatially extended involves changes of gait or continuous modulation of a nature of the body are neglected. Applying this formal- single behavior. At the same time, there are conditions ism to a perfectly sinusoidal body wave of low amplitude in which organisms display very uncoordinated locomo- (A) and short wavelength (λ ≪ L) (Fig. 4A), Gray et 28,29 tion or else fail to make progress altogether. These may al. were able to derive an expression relating vslip to correspond to environments that are not usually encoun- vwave: tered in an organism’s natural habitat, or – more often in the geneticists’ laboratory – to mutants that lack an B +1 v = v , (12) essential protein. The investigation of biological forms slip wave KB +1 of undulatory locomotion across different physical envi- ronments dates back to the early 20th century22,23,24,29,32 where B =2π2A2/λ2. and is playing an increasingly important role in genetics In general, for a given locomotion waveform, the de- and in neuroscience31,33,34,35. gree of slip depends only on K. Relating this back to the Until now, we have limited ourselves to Newtonian case of locomotion in a rigid channel where we effectively fluids, which are fully described by the viscosity of

6 ContemporaryPhysics,toappear 3 APPLICATIONS TO BIOLOGICAL SWIMMERS AND CRAWLERS

lated microorganisms. Indeed, this field already occu- A B pies a vast literature (see36,37,38 for reviews). Here we present only a brief overview of two examples of molec- ularly driven undulations at the subcellular level: flag- ella mediated swimming in spermatozoa and in bacterial C D cells. We then proceed to discuss locomotion in non- Newtonian environments, using small nematode worms as a case study.

3.1. Flagellated microswimmers FIG. 4: Assessing the validity of Eq. (12) with a physics sim- ulator (see text for details). A. The analytic result of Gray et al.29 is an excellent approximation for an undulator with Bacteria and spermatozoa self-propel by means of or- a perfectly sinusoidal wave, a wavelength that is short rela- ganelles called flagella. A flagellum is a long filament that tive to the body length and of a low amplitude. B. A purely protrudes from the cell and provides propulsion. How- sinusoidal idealization of the C. elegans crawling wave. The ever, the flagellar motion of eukaryotic cells (like sperm) same equation handles such a wave reasonably well but with is significantly different from that of bacteria. While a sperm’s flagellum makes sinusoidal undulations in 2D, small errors due to the increased amplitude and wavelength. bacterial flagella make helical undulations. C. A more realistic C. elegans crawling waveform in which Eukaryotic flagella consist of a core of two microtubules the wavelength increases towards the tail introduces signifi- (long, tube shaped polymers) surrounded by nine micro- cant errors. D. A realistic C. elegans swimming waveform in tubule doublets, arranged to form a circle. Each micro- which the wavelength is greater than the body length is very tubule is as long as the entire flagellum and typically ex- poorly described by the analytic approximation. tends a few body lengths. The flagellum of, for example, a sea urchin spermatozoon undulates at about 50Hz, with a nearly sinusoidal waveform39. This bending is achieved by what is essentially a distributed molecular motor con- the medium, but many low Re swimmers in fact move sisting of the microtubules themselves. Powered by the through complex fluids or other ‘soft’ environments. For hydrolysis of ATP, pairs of microtubule doublets slide rel- slender bodies in a Newtonian fluid, the ratio of drag ative to each other40. In order to generate the alternate coefficients K is fully determined by the geometry and bending characteristic of a sine wave, this sliding motion cannot exceed 2. In contrast, in non-Newtonian envi- is resisted by local inter-doublet links at specific points ronments, K (if and when it is well defined) is both a along the flagellum. By propagating these zones of slid- function of the geometry and of the medium. Strictly ing and linking along the flagellum, the sinusoidal wave speaking, viscosity is not defined in non-Newtonian flu- is generated and propagated. The control of this process ids, since the linearity of Eq. (1) is violated. Thus even is chemically “hard wired” into the system. the Navier-Stokes equation is not applicable. For exam- Despite generating force according to similar underly- ple, the fluid may have some non-trivial structure, it may ing physics, the control of a bacterial flagellum is com- have energy storage capacity (e.g., elasticity), or perhaps pletely different37,40 and unique to bacteria. Unlike eu- the properties of the fluid may depend on the speed with karyotic flagella, bacterial flagella are passive fibers inca- which it is deformed. Of these cases, visco-elastic flu- pable of active bending. The solution they use instead ids (and visco-elastic approximations of gels) are prob- involves helical waves. The bacterial flagellum itself is a ably the most relevant to biological swimmers. Slender thin filament, grossly similar to the eukaryotic one. How- body or resistive force theories can both be straightfor- ever, rather than the microtubules of eukaryotic flagella, wardly extrapolated to model visco-elasticity when the bacterial flagella consist of proteins called flagellin ar- elastic properties of the fluid can be approximated by ef- ranged to form a hollow cylinder. Subtle properties of fectively stronger resistive drag coefficients in the normal the flagellin’s physical structure mean that the resulting direction, thus increasing the ratio K29,98. In what fol- flagellum is not straight but helical, which is integral to lows, we loosely call such extensions to Newtonian fluids, their function in locomotion. The entire helical flagellum anomalous K environments. is connected via a sharply bent construct called a hook to To the extent that slender body theory (and the sim- the shaft of a rotational motor in the cell’s membrane38. plified slip formulation thereof) may be extrapolated in This molecular motor bears a striking resemblance to an this way, all of the above logic still holds, except that electrical stepper motor and is powered by the flow of now the slippage parameter K is no longer determined protons across the cell’s membrane. Thus the flagellum solely by the geometry of the swimmer. acts exactly like a boat’s propeller attached to the drive shaft of a motor. How does this mechanism generate thrust? In the case of eukaryotic flagella, the posteriorly directed compo- 3. APPLICATIONS TO BIOLOGICAL nents of the motion add up but the sideways components SWIMMERS AND CRAWLERS counter balance, thus generating progress in a direction parallel to the long axis of body (axially) as described The hypothetical cylindrical organism described above in Section 2.3.3. Now, to understand bacterial flagella, is a good approximation for a vast range of different think of each part of the helix as a small tilted rod41. micro-swimmers, from bacteria to microscopic larvae or When the rod rotates within the spiraling helix, it exerts worms. Of these, the best studied examples are flagel- an axial and sideways force on the fluid leading to for-

7 ContemporaryPhysics,toappear 3.2 Undulations of microscopic worms 3 APPLICATIONS TO BIOLOGICAL SWIMMERS AND CRAWLERS ward propulsion and an overall torque. The torque then as a wedge, forming a channel through the grains22,29. induces a counter rotation of the cell body. Motion in the normal direction would require many more On a final note, we mention the interaction that occurs grains to be displaced. Although the resistive forces of when multiple flagella surround a bacterial body and in- these media cannot strictly be called drag, their behav- teract to generate complex motion. This relies on the ior is nonetheless frequently represented in terms of local fact that the motor driving each filament can spin both resistance coefficients ck and c⊥. In many cases the non- clockwise and counterclockwise. When spinning coun- Newtonian properties of these fluids are sufficiently well terclockwise, all the flagella rotate synchronously, form- modeled using this approach. ing a bundle that propels the cell forwards. When spin- ning clockwise, however, the filaments fail to synchro- nize and each one ends up pushing the cell in a different direction42,43. Thus the bacterium goes nowhere, but Experiments does undergo some random rotation. On a final note, we mention the interaction that occurs Given the fundamental importance of environmental when multiple flagella surround a bacterial body and in- properties to undulatory locomotion, it would be ben- teract to generate complex motion. This relies on the eficial to experimentally determine the relevant proper- fact that the motor driving each filament can spin both ties of the swimmer’s environment (and in the case of clockwise and counterclockwise. When spinning coun- the worm, particularly those of agar gels). This would terclockwise, all the flagella rotate synchronously, form- help to determine how valid the slip formalism is for ing a bundle that propels the cell forwards. When spin- such media and to facilitate the development of quanti- ning clockwise, however, the filaments fail to synchro- tative models of an organism’s locomotion. For an ideal nize and each one ends up pushing the cell in a differ- Newtonian fluid it suffices to know the viscosity, which ent direction42,43. Thus the bacterium goes nowhere, can be easily measured by even the simplest rheometers. but does undergo some random rotation. In a typical Given the fluid viscosity and the dimensions of the or- environment, a bacterium will switch between the two ganism, good estimates of the drag coefficients can be modes of flagellar rotation and alternately rotate and obtained. If greater accuracy is required then the full swim in a so-called tumble and run motion. Bacterial Navier-Stokes equations may be used. However, even in sensory mechanisms modulate the frequency of tumbles, Newtonian environments, complications can arise. For such that when a bacterium is in a favorable environment example, nonuniformity in the medium may result in in- (rich in nutrients) it will turn often and tend to stay in homogeneous viscosity. This is especially a problem for the same vicinity, whereas if conditions are less favor- very small swimmers, where inhomogeneities on scales able, runs will be longer and the bacterium will undergo of microns or tens of microns may locally influence the a biased random walk in search of food44. motion. Additional complications arise when dealing with visco-elastic fluids, gels and suspensions of particles. While even complex fluids can be well characterized using a modern rheometer, the quantities measured (e.g., the 3.2. Undulations of microscopic worms elastic and loss moduli as a function of frequency) are not behaviorally relevant. In principle, advanced sim- To demonstrate some of the physics of simple non- ulation techniques could be applied to predict physical Newtonian environments, we focus in what follows on a forces from rheological properties, but this would be an small nematode worm called Caenorhabditis elegans. C. extremely difficult undertaking. elegans is a leading model organism for biologists and as It turns out that a simple representation of fluid be- such is grown extensively in laboratories, where it is cul- havior in terms of local drag coefficients ck and c⊥ is tured on the surface of agar gels. With a length of 1mm sufficiently accurate for many types of study, if only we and up to 2Hz undulation frequency, the worm can have could measure their values. One of the complicating fac- a Re of about 1 in water – approaching the upper limit tors in such an experiment is the small size of the organ- but still within bounds for a low Re treatment. isms under investigation. Nonetheless, some innovative C. elegans worms crawl on the agar surface with very approaches have been used. little slip, suggesting that the value of K is high. Close By dropping small diameter wires through various inspection reveals sinusoidal tracks left by the worms as Newtonian fluids, Gray and Lissmann29 were able to ver- they move. Each track is actually an indentation (or ify the expected value of K ≈ 1.5. By performing the groove) in the surface of the gel, and helps to explain the same experiment in agar gel, they were able to show that lack of slip. Although the worm’s mass is negligible, a K was significantly greater (though no value is reported). thin film of water forms around it and the resulting sur- In order to quantitatively estimate the forces experienced face tension presses it strongly against the gel surface45. by a small nematode crawling on agar, Wallace placed As the worm moves forwards, it overcomes the gel’s yield small glass fibers (with dimensions similar to the nema- stress and breaks the polymer network with its head. tode) on the agar surface and measured the force required This allows the rest of the body to slip forwards more to pull them along, thus estimating the propulsive force easily. Motion normal to the body surface is strongly re- exerted by the nematode47. He also performed experi- sisted because such motion would require further break- ments where the thickness of the water film around the ing of the polymer network, over a much larger area. For worm was altered, thereby modulating the forces act- a sufficiently stiff gel, the net result is functionally quite ing on it and resulting in different locomotory behavior. similar to locomotion in a solid channel and allows K to More recently, Lockery et al.34 have developed an alter- be significantly larger than in the Newtonian case29,46. native strategy of imposing K = ∞ by manufacturing A similar effect occurs when an organism moves micro-fluidic chips with tiny channels of pre-determined through a granular medium like soil. Again the head acts shape filled with water. While direct measurement of

8 ContemporaryPhysics,toappear 4.1 K estimation 4 SIMULATIONS physical forces in fluid environments would be preferable, with the center of mass trajectory. Thus we are effec- the ability to effectively set these parameters to known tively performing a gradient descent on Fnet and τnet in values is a powerful tool. order to find the correct trajectory. Within the slip formalism, the progress of the organism does not depend on the actual values of ck and c⊥, just on the ratio K. Given that internal forces are not modeled, 4. SIMULATIONS it is simply assumed that they are sufficient to produce the recorded changes in body shape. Thus, for a known Undulatory locomotion of roughly cylindrical organ- K, this kind of simulator can simply solve the equations isms at low Reynolds number is well suited to imple- of motion. Alternatively, if K is not known, but the co- mentation in computer simulations. In particular, when ordinates and rotation of the body shapes are known, body bending only occurs in a plane, as in C. elegans, a then a simulator like this can also be used to estimate 2D simulation is sufficient to capture the dynamics, mak- K. Thus, this simulation approach can be used to esti- ing the task simpler and less computationally expensive. mate environmental properties without any direct mea- In what follows we describe a number of simulators that surement of that environment (except the visual record- can be used to study worm locomotion in a variety of ing of the motion)31. media.

4.2. The physics simulator applied 4.1. K estimation As mentioned above, a physics simulator is particularly As we have already seen, in low Re environments, the valuable for computing the progress of a long and slender trajectory of a body is entirely determined by its se- undulator in a Newtonian or anomalous K environment. quence of shapes and the properties of the environment. In addition, such a tool can also be used to address basic One approach then, is to use the locomotion traces of questions in slender body theory or in low Re physics. different bodies to estimate the properties of the environ- Here we briefly describe three such examples. ment. If the model of the environment is approximate, such as reducing the description of the environment to two drag coefficients, then a simulation approach can also 4.2.1. Limits of the slip formulation serve to assess the validity of the theory31. If valid, the simulator can then be used to assess further approxima- 28,29 The physics simulator as described above uses a ba- tions, such as the slip formalism in Eq. (12) . sic result of slender body theory, namely that motion lateral and tangential to the body experience different drag forces (and extrapolated to arbitrary ratios of these Solving equations of motion at low Re forces). However, it makes no assumptions about the configuration of the body. One can therefore ask31 (i) whether the results of slender-body physics are valid ap- In a typical, high Reynolds number physics engine, the proximations of, say, the forces experienced by worms net force on a body results in acceleration according to moving on agar and (ii) whether further simplifications, a = F/m. At low Re, the very small mass will lead to e.g., as given by Eq. (12), are valid. The latter would very large accelerations, leading to a stiff system requir- require a simulation of artificially generated body shapes ing very short time steps. In the limit of m → 0 and in a pre-specified environment and the comparison of for F 6= 0, we will have the numerically problematic sit- the simulated progress with the theoretical prediction. uation of a → ∞. In the “real world”, this means that For example, Fig. 4 shows that three of the assumptions the velocity of the body will always be at steady state, needed to derive Eq. (12) break down for worm-realistic at which the net force and similarly net torque are zero. skeletons. These are first, that the locomotion wave is Solving the equations of motion is therefore tantamount sinusoidal; second, that the wavelength λ is short com- to satisfying these conditions. pared to the body length L; and third, that the ampli- For example, one might begin by comparing the body tude A is sufficiently small so that the wavelength λ is shapes at times t(i) and t(i − 1), assuming zero progress. similar to the corresponding arclength along the undu- Given the model of the environment, one can then calcu- lating object’s body. Such results, while straight forward late the reactive environmental force and combine these to generate, nonetheless offer a quantitative handle on to obtain the net force Fnet and torque τnet acting on the commonly made approximations in the field. body. For our simplified model of Newtonian or anoma- lous K environments, one need simply decompose the velocities of each point into their normal and parallel components and calculate the corresponding drag forces. 4.2.2. Purcell’s scallop in non-Newtonian environments Based on the directions and magnitudes of these vectors, we can now apply a small displacement and/or rotation One interesting question is to what extent the scallop (in the direction of −Fnet and −τnet) and recompute the theorem applies to more complex environments. Review- reactive forces to get the new resultant. This process can ing the argument in Sec. 2.2, it is easy to see that the be iterated, accumulating small displacements/rotations extension of the physics to variable K does not intro- until the zero net force and torque conditions are met duce any time asymmetry into the governing equations, (within a specified tolerance), at which point we com- and so the scallop theorem should hold. The physics bine the final displacement and rotation δx, δy and δθ simulator above is well suited to simulating scallops in

9 ContemporaryPhysics,toappear 4.2 The physics simulator applied 4 SIMULATIONS media with different values of K, so this generalization component. Equation (13) guarantees only non-negative of the scallop theorem can easily be verified numerically contributions to energy dissipation, and easily lends itself and indeed holds true, as would be expected. to numerical minimization. Figure 5 demonstrates this approach with one such example. The body and environment in this example 4.2.3. The variational principle and Helmholtz’s Theorem are represented in the same way as described in Sec. 2.3.1. For this example (and indeed for a variety of wave- forms and actual movies of C. elegans locomotion), the As described above, the conventional way to determine minimum energy dissipation trajectory agrees with the the dynamics of objects in Newtonian fluids requires a so- solution of the equations of motion, to within the dis- lution of the Navier-Stokes equations, but this can prove cretization step of the energy sweep. The simulation also to be computationally difficult. To complement this ap- demonstrates the intuitive result that the landscape (as proach, attempts have been made to tackle fluid dynam- far as explored) is smooth and has only a single mini- ics, or at least low Reynolds number Newtonian fluid mum. This approach, while far from an analytic proof, dynamics, from a very different perspective: that of min- demonstrates the potential application of such simula- imization of energy dissipation. Put simply, if the evo- tors. Specifically, the set of simulations in this example lution of the system is too difficult to calculate, one can suggests that slender body motion, whether passive or put forward an arbitrary path, and then use variational self-propelled, in any environment characterized by ck methods to find the path that minimizes the energy dis- and c⊥, will obey the principle of minimum energy dissi- sipation of the system, be it heat dissipation in an elec- pation. trical circuit48 or at least in principle, due to drag forces in a fluid49. If this so-called principle of minimum en- ergy dissipation was true, this path would then satisfy the equations of motion. In fact, minimum energy dissi- 4.3. Modeling worms pation has been proved anecdotally, in very simple cases, but in other cases it has been found not to hold at all. A physics simulator like the one described above is Thus, while appealing and intuitive, it appears not to be useful in cases where the locomotion waveform can be a general principle. Having said this, it is still an open recorded or pre-determined, but intentionally avoids the question to try to understand under what conditions the question of how an object or organism generates its un- principle applies, typically because of the potential for dulation wave. In general, the shape of the body is de- computational advances in difficult problems where the termined by a combination of (i) internal (molecular or equations of motion are particularly challenging to solve. muscle) actuation forces, (ii) physical properties of the What then, is the status of this problem in fluid dy- body and (iii) external forces from the environment. namics? A theorem due to Helmholtz and Korteweg Consider again our friend C. elegans. Conveniently, the proved the minimum energy dissipation principle in the worm’s neuro-anatomy essentially limits it to bending in case of a bounded volume filled with a viscous fluid where 56,57 the boundaries are moving with a well specified velocity, two dimensions . Nematodes, like other worms, lack 50,51 any form of rigid skeleton and instead rely on the antago- and in the limit of negligible inertial forces . Only nistic forces of their elastic outer casing (cuticle) and the very few studies have attempted to tackle other bound- hydrostatic pressure within their body cavity to maintain ary conditions, such as spheres or ellipsoids immersed 52,53,54,55 their shape. Thus, at the simplest level, the body can be in a Newtonian fluid and moving under gravity . represented by an elastic rod or cylinder with some in- The more general case remains open. ternally controlled bending along the rod. Alternatively, One interesting case for which results have not been more elaborate and detailed models are possible that al- derived is that of a self-propelled body, i.e., with a time- low for more accurate biophysical and biological ground- changing configuration. Another interesting extension is ing. to non-Newtonian media. We restrict the discussion to In the nematode literature, the canonical example is undulatory swimming through Newtonian and anoma- a model of C. elegans locomotion due to Niebur and lous K fluids at low Reynolds number, where despite be- 46 ing intractable analytically, it is possible to study the Erd¨os , which is based on exactly these principles. Their problem using numerical models and simulations. An 2D model approximates the body as an elongated rect- advantage of a simulation approach is that it can be angle represented by two rows of points along its outline. applied to arbitrary artificial body skeletons, as well as Adjacent points are connected by springs (representing body shapes extracted from movies of actual swimming the cuticle) and are pushed apart by pressure forces. This (where both the shape and coordinates of the body are body also behaves effectively as an elastic rod with a known) and for a variety of (Newtonian as well as non- characteristic persistence length. Note that until now, Newtonian) drag coefficients. For a long and slender we have not evoked any low Re considerations (except body in low Re, motion in any environment that can be that we have not endowed our ‘rod’ with any mass). described by drag coefficients c and c will be subject The low Re physics does come in when solving for k ⊥ the changes of shape and position of the body. In the to energy dissipation of the form above model, points are able to move independently un- T der the action of muscle forces, and modulated by the 2 2 environment. To solve the model, one can invoke once E = [v⊥(t,s)c⊥ + vk(t,s)ck] ds dt (13) 0 more the slender body theory and approximations due Z Z to Gray et al.28,29, and decompose the forces into tan- over a cycle time T , where v⊥(t,s) is the normal com- gential and normal components Fk and F⊥. Much like ponent of velocity at time t and at position s on the the physics simulator described above, the zero net force body surface, and vk(t,s) is the corresponding tangential condition can be applied to calculate the velocity of each

10 ContemporaryPhysics,toappear 5 SENSORY INTEGRATION AND CLOSED LOOP CONTROL

A

0.2

0 B 5 Y (mm) −0.2 4

−0.4 −0.2 0 0.2 0.4 0.6 max 3 X (mm) E/E 2 C 1.001

1.0008 1 −1 1.0006 0.2 max −0.8 1.0004 0.1 E/E −0.6 1.0002 0 −0.4 −0.1 1 0 0.5 1 1.5 2 2.5 ∆ µ −0.2 −0.2 ∆ Y (µ m) −5 X ( m) ∆ ω (rads) x 10

FIG. 5: The principal of minimum energy dissipation correctly predicts the displacement and rotation of a self-propelled worm in a 2D anomalous K medium. Simulations confirm that the solution of the equations of motion coincides with the minimum energy dissipation. A. An undulating body with sinusoidal shape moving according to the equations of motion. Red arrows represent the drag force acting on each point. The “CoM” (blue circle) moves with velocity V (blue arrow) such that the net drag force is zero. B, C. The corresponding energy dissipation landscape based on a sweep of displacements (x,y) in a plane (B) and rotation in the plane (C). In both plots the minimum energy dissipation is marked in red on the landscape. The motion obtained from the equations of motion (shown in A) is indicated by a blue line. The results agree to within the step size.

point along the body. Specifically, we can write write generates rhythmic patterns of activity. Neural circuits Fdrag,i = −civi = −Finternal,i along each direction i, and that generate rhythmic patterns of activity even when since we already have our internal forces, it remains to completely severed from the rest of the body are dubbed 58,59,60 extract the speeds along each direction vi = Finternal,i/ci. central pattern generating or CPG circuits .

The fact that a CPG functions in isolation suggests 5. SENSORY INTEGRATION AND CLOSED that it can also be modeled in isolation. Indeed, tradi- LOOP CONTROL tionally, most animal motor control models (locomotion included) tended to be limited to a bottom-up model To complete the model of an undulating system, one of internal neuronal or neuro-muscular dynamics. These may also include the active elements that induce and con- models are then complemented by top-down models of trol the bending of the body. This may be the molecular the physical aspects of locomotion (e.g., the aerodynam- biophysics underlying force generation in flagellar mo- ics of flight, the mechanics of legged locomotion and so tion, or a much higher level model of brain and body on). The underlying assumption here is that, to a first control in an undulating animal. Either way, one can approximation, the neuronal control can be treated as a think of a motor control system as consisting of a con- stand-alone control unit and hence decoupled from the trol signal that drives actuators or motor elements to physics of the body and environment. control the shape of the body, subject to environmental forces. In animals, the control signal is generated by the When then does the physics matter sufficiently to jus- nervous system; the actuators are muscles; and feedback tify an integrated neuro-mechanical model? In what fol- messages are generated either by the body (signals that lows, we describe two examples of swimming undulators: sense the state of the muscles as well as the orientation the lamprey, and C. elegans. In the lamprey, the vast and shape of the body) or by other sensory perception majority of modeling has focused entirely on the isolated (vision, etc.). For example, when we lift a heavy object neural control of locomotion. Recent exceptions are be- or push against a wall, the body senses a strong resis- ginning to highlight the role that physics plays in the tance from the environment and can relay this to the locomotion and are perhaps opening new avenues of in- nervous system which, in turn, can alter or modulate the vestigation of this classic model system. In C. elegans, neural control. Thus, in this example the neural signal the low Re physics has long been recognized to play a can be thought of essentially as a centrally controlled significant role, but models have still tended to decouple system with some feedback. It turns out that the vast the nervous control from the mechanics. Here too, exper- majority of animal motor control systems contain such iments and models are moving increasingly in a direction neural “control boxes”. Since most muscle behavior con- of an integrated study of the entire locomotion system, sists of rhythmic motion, the underlying neural control bottom-up and top-down.

11 ContemporaryPhysics,toappear 5 SENSORY INTEGRATION AND CLOSED LOOP CONTROL

5.1. A high Re example: Lamprey swimming A The lamprey is a primitive, eel-like aquatic vertebrate LR that can reach up to one meter in length. It swims by propagating lateral undulations with increasing ampli- tude along its body (from head to tail). Lampreys have a brain and spinal cord with a relatively complex nervous system. The spinal cord serves to relay information be- B tween the brain and body, but is also capable of complex EE dynamic behavior. In the 1980s it was shown that the basic neural activity pattern responsible for locomotion could be produced by the isolated spinal cord61, with- out any input from the brain or from sensory pathways. This became the canonical example of a vertebrate CPG II circuit62,63,64,65,66. It was further shown that the lamprey CPG consists of a sequence of semi-independent neural oscillators that are capable of independent pattern generation, but are coupled to each other in such a way that the correct C CIN CIN phase lag between adjacent units is preserved67. Thus, the CPG circuit can generate stable rhythmic activity (at swimming frequencies) that propagates along the spinal cord, and where each ‘segment’ along the body exhibits anti-phase oscillations between the left and right sides of the body (so that when connected to the muscles, one MNEIN EIN MN side would contract when the opposite side relaxes). In most CPG circuits, the prevalent model of anti-phase oscillators relies on the half center oscillator66,68, which consists of two units that are coupled via reciprocal inhi- bition (as shown schematically in Fig. 6A). These units typically represent separate pools of neurons (as they do LIN LIN in the lamprey). External driving forces and/or internal dynamics modulate the activity to generate rhythmic, out of phase activity. FIG. 6: Schematics of neural oscillators for motor control. Early models of the lamprey spinal cord relied on ex- A. The half-center oscillator is the simplest central pattern actly these principles, and have led to important ad- generating circuit (excluding intrinsically oscillating neurons) vances, both in the study of the lamprey and other ner- and consists of two mutually inhibiting neurons. An addi- vous systems62,71,72, and in our understanding of coupled tional requirement is that the neurons can be released from, oscillator systems in nonlinear dynamics64,65,73. While or are able to escape from, the inhibition. An example of a the lamprey swimming CPG circuit (Fig. 6C) looks release mechanism is synaptic fatigue, where the inhibition rather complicated, the motif of reciprocal inhibition that wears out over time. B. Simplified nematode oscillator net- is required for generating oscillations is not very differ- work (adapted from Ref. [69]). (E)xcitatory neurons innervate ent from the minimal circuit in Fig. 6A. The circuit in- the (I)nhibitory neurons which in turn relay the inhibition to cludes both excitatory and inhibitory pools of neurons the opposing excitatory neuron. C. The lamprey segmental that innervate motor neurons that connect to the mus- oscillator (adapted from Ref. [70]) is more complex, but func- cles. Excitatory interneurons (EIN) provide the excita- tions along similar principles. It consists of motor neurons tion to switch the circuit on; contralateral inhibitory in- (MN) along with three classes of interneuron (IN), with CIN terneurons (CIN) provide the cross inhibition, and lateral providing the contralateral inhibition. inhibitory interneurons (LIN) help to terminate contrac- tion of each side74. This principle is demonstrated in an elegant robotic instantiation by Auke Ijspeert et al.75, that builds on the integrated neuro-mechanical model of of the body. In principle, this approach, if further com- lamprey swimming due to Ekeberg74. plemented by an understanding of the fluid dynamics, Over the years, as models of the lamprey motor con- should lead to further progress in understanding how the trol have become more refined and sophisticated, research lamprey and other systems exploit the physics of the en- has increasingly focused on modeling the modulations of vironment to modulate their internal neural dynamics the swimming rhythms by input from the brain, from and motor control. sensory neurons and by the local action of chemicals (so- called neuromodulators) and even modulation by propri- oceptive inputs, but of these, the role of sensory effects are least understood76. To shed some light on the ef- 5.2. A low Re example: C. elegans fects of sensory feedback, Simoni and DeWeerth77 have recently introduced a neuromechanical lamprey model Imagine a system in which the neural control is driven with sensory feedback to study synchronization effects directly by sensory signals (such as reflexes). Such a neu- between the CPG circuit and the mechanical resonance ral system would completely fail in the absence of a body.

12 ContemporaryPhysics,toappear 5.2 A low Re example: C. elegans 5 SENSORY INTEGRATION AND CLOSED LOOP CONTROL

Surprisingly, such appears to be the minimal model de- circuit bears a distinct resemblance to the simple half- scribing C. elegans locomotion31,46,78. We have already center oscillator in Fig. 6A, except that the inhibition is briefly described some of the worm’s means of locomo- mediated via distinct inhibitory neurons. Nonetheless, tion and its exploitation of the environment to generate for a combination of reasons, questions have been raised thrust (see Secs. 3.2 and 4), but here we focus on the about the ability of this circuit to support a CPG net- internal (neural) control of the worms’ undulations. work. This tiny nematode worm has an invariant anatomy One may therefore ask what would be a working model that consists of a mere 959 cells in the adult. Of these, of the locomotion in the possible absence of a CPG mech- about a third (302 cells) are nerve cells or neurons. These anism. One interesting conjecture is that the worm may make up a simple nervous system that is distributed use a sensory feedback mechanism as the primary driver throughout the animal. Remarkably, the nervous system of its neuronal oscillations. Weight is given to this hy- appears invariant. In other words, the connectivity dia- pothesis by several models of the worm’s locomotion that gram or neural circuitry is the same across worms. This are grounded in its anatomy and successfully produce connectivity diagram was meticulously mapped out56,57 oscillations by means of mechanosensory feedback79,80. and is the starting point for understanding the worm’s How would such a mechanism work? We begin by dis- behavior. cussing the worm’s so-called crawling behavior (short Locomotion is the worm’s principal motor activity and wavelength undulations observed on agar). This was ad- mediates everything the worm does, from foraging to dressed by the first model of C. elegans locomotion due mating. Thus, it is not surprising that of its 302 neu- to Niebur and Erd¨os46. Their model of the worm’s body rons, around 200 are involved in locomotion. Of those, was introduced in the Sec. 4.3, but they also proposed a about 50 are directly responsible for the generation of simple model nervous system. undulations and their backward propagation down the This model involved a hypothetical CPG circuit in the body, i.e., the generation of forward locomotion. (The worm’s head, allowing it to generate a sinusoidal trajec- worm can also move backwards, propagating waves from tory that was then preserved by strong lateral resistance, tail to head, but this behavior relies on a different set of due to a very stiff groove (high K). The worm’s shape neurons). To put these numbers in perspective, the hu- was therefore determined by the shape of the groove, and man brain contains of the order of 1011 neurons and an the model nervous system was then responsible for gen- insect such as the tiny fruitfly has about 105. Thus, while erating muscle contraction down the body, pushing the the nervous systems of most animals (including lamprey worm forwards. To control the rhythmic muscle activity and even the fruitfly) rely on large populations of neu- and rhythmic underlying neuronal activity, the neurons rons to perform processing, in C. elegans single neurons in the model were endowed with so-called stretch recep- must do the same job. tors, or mechano-sensitive ion channels that activate an When nematodes are placed in environments with dif- ionic current in response to stretch. The current then ferent viscous, visco-elastic or gel properties, the pa- activates that neuron, and local muscle contraction is rameters of their locomotion wave change quite signifi- induced. If muscles on one side of the body are maxi- cantly. For instance on agar gels, the worms exhibit slow mally contracted, the tissue on the other side is maxi- (∼0.5Hz) short-wavelength undulations (with the body mally stretched. Crucially, the model assumes that the length spanning about 1.5 wavelengths), whereas in wa- stretch receptor signal is transmitted some distance for- tery environments, they appear to thrash, i.e., undulating ward from the site of the stretch: The existence of a set much more quickly (at ∼2Hz) with a wavelength of just distance between the stretch detection and the ensuing under twice the body length. In fact, the entire range contraction can result in the propagation of the undu- of intermediate behaviors can also be obtained in envi- lation in the desired direction (from head to tail). This ronments with various degrees of visco-elasticity31. In integrated neuromechanical model was able to reproduce principle, the observed change in behavior could be the the crawling behavior on agar if the groove was assumed result of internal changes to the neural circuit (e.g., by to be sufficiently stiff (with very high K). The main lim- neuromodulators, or sensory pathways that activate dif- itation of this model was the fact that it only worked for ferent sub-circuits). Alternatively, the entire change in high K, and therefore failed to produce any oscillations behavior may be down to changes in the physical inter- in a virtual environment equivalent to water. At the time action between body and environment. however, it was believed that the worms swimming in wa- ter and crawling on agar were separate locomotory gaits To shed light on this problem requires a closer look at 56 the locomotion nervous circuit of the worm. Unlike verte- (like a horse’s trot and gallop) . One might therefore brates, the worm does not have a spinal cord, but it does conjecture that the two behaviors would involve separate have an analogous ventral cord which contains the loco- neural mechanisms. motion motor neurons, and so we shall focus on it. First, So how does an organism with such a limited neural one may look for evidence of characteristic motifs along circuit adapt its locomotion waveform in environments the ventral cord that are expected to generate rhythmic with drastically different properties? The fact that the activity, such as reciprocal inhibition. This typically re- worm is capable of a continuous range of behaviors sug- quires breaking up the circuit into small units (analogous gests that it relies on a continuous modulation of a sin- to segments or vertebrae in the lamprey). Unfortunately, gle mechanism grounded in a single circuit. The most even this first step is not a trivial task, since the nervous parsimonious explanation would involve an effectively circuit of the worm contains very different numbers of unchanged neural system, whose activity is modulated neurons on different sides of the body, and so is not nat- solely by the physics of the environment. This is not un- urally divisible into repeating units. Based in part on likely in a low Re environment. In fact, passive flagella data from related nematodes, in which direct recording of bacteria also exhibit shape changes when the viscos- from neurons is possible, a model of the generic nema- ity of their environment is increased81, and even more tode segmental oscillator (Fig. 6B) was proposed69. This pronounced effects can be observed in the waveform of

13 ContemporaryPhysics,toappear 5.2 A low Re example: C. elegans 6 APPLICATIONS TO MOBILE MICRO- AND NANOROBOTICS a sperm’s flagellum (where active bending occurs along most living systems combine central control with sensory the length of the filament)82. This suggests that a modu- modulation83,84,85 and where advantageous, such systems lation of waveform by physical forces is worth exploring. will exploit the physics of the environment. Indeed, more Within the context of a neural system, one would ex- and more, examples are being studied in which sensory- pect a mechanism driven by sensory-feedback to be more motor control takes over when the CPG activity is some- naturally suited to direct modulation by the physics of how disabled85. the environment than a CPG controlled one. Further- more, one would expect that without the constraints of an internally clocked system, the modulation could take place over a wider dynamic range. Such a mechanism 6. APPLICATIONS TO MOBILE MICRO- AND would need to explain changes in frequency, amplitude NANOROBOTICS and wavelength of undulations31. It is not surprising that an increase in mechanical load In this paper we have focused on a few examples of on the body and muscles (with increased viscosity or low Re undulatory locomotion, but the principles, the- visco-elasticity) will result in slower contraction. To ex- ories and techniques can all be applied to many other plain the corresponding decrease in wavelength and am- situations. There are countless organisms that use undu- plitude, it is convenient to imagine the body initiating latory locomotion similar in form to that presented here, an undulation from a straight line configuration. The over a range of Reynolds numbers and spanning some 6 higher the environmental resistance, the harder it is for orders of magnitude in size. Moving to high Reynolds the body to bend. This change is most pronounced in number, for example, means that certain principles (like the middle of the body, as bending here requires signifi- the scallop theorem) no longer apply but others (like the cant movement of the head and tail. The head, which is requirement of symmetry breaking) still do. Here we will strongly activated and has one free end, will react first. begin by introducing some interesting biological exam- At some point the head will be sufficiently bent that the ples (taken from our own bodies) of systems that, while stretch receptors trigger and the direction of bending is incapable of locomotion per se, bear similarities to what reversed. The mid-body meanwhile is still trying to bend we have covered so far. This will be followed by some in the first direction and will only reverse after some de- examples of biologically inspired robotic applications of lay. (This assumes that during locomotion, a straight undulatory locomotion. worm is unstable and will tend to bend.) As the vis- It is perhaps not surprising that biology has applied the cosity increases, the phase lag between adjacent parts of principles of undulatory locomotion to the task of moving the body will increase, leading to a shorter and shorter substances around. One example of this is peristalsis, the wavelength. mechanism by which food is moved through our digestive Consider now the average body curvature which in- tract86. Imagine taking a swimmer and attaching it to creases with viscosity or visco-elasticity. Assuming that a solid object at one end. What we are left with is es- stretch receptor signals are integrated over some length sentially a pump. Our intestine, while anchored in place, of the body, the average curvature will be related to the generates waves of muscle contraction which pump semi- wavelength of the undulation99. Specifically, the higher digested food through our system. The smooth muscle of the curvature, the shorter the wavelength. When the our intestine can only contract inwards (reducing the ra- wavelength is short, the stretch receptor integrates over dius), but by propagating coordinated waves of activity a significant fraction of the wavelength. Thus while some the food is propelled in one direction by the leading edge parts of its receptive field will be highly curved, other of the wave in much the same way that our hypothetical parts will be nearly straight, or even curved the other organism in Sec. 2.3.2 generates thrust. way. Conversely, when the wavelength is long, the curva- Another type of biological pump can be found, among ture will be more homogeneous along the receptive field. many places, in our trachea (wind pipe). The inside of Thus, while the level of integrated curvature required to our trachea is lined with organelles called cilia, which cause a reversal of bending will be the same across the are genetically very closely related to eukaryotic flagella, different environments, the peak curvature will be greater and possess the same 9+2 microtubule structure87. Com- when the wavelength is shorter. pared to flagella, cilia are generally shorter and are found The above reasoning was used in an integrated in greater numbers. While a flagellum generally produces neuro-mechanical model of nematode locomotion that thrust in the direction of its long axis (like a propeller), reproduced the swim-crawl transition across different cilia generally produce thrust perpendicular to their long environments78. This model is grounded in the known axis (like oars). Their asymmetric beating pattern88 is neural circuitry (a variation on Fig. 6B), and does not such that they minimize drag on the forwards stroke and contain a CPG either in the head or along the body. The maximize drag on the backwards stroke, thereby generat- only parameter changes required to modulate the loco- ing thrust along the surface from which they protrude. In motion are to the drag coefficients ck and c⊥. The key so doing, they move mucus and dirt away from our lungs ingredients of the model are bistability in the motor neu- ready for expulsion. In fact, cilia perform other roles as rons (leading to instability of straight line configurations) well. In many cases they act as sensory organs87 (such and sensory feedback via distributed stretch receptors. as the cilia in our ears), but that is outside the scope of Having reviewed a system that can rely entirely on this paper. Motile cilia are also found on some micro- sensory (proprioceptive) feedback to drive and modulate organisms like protozoa (unicellular eukaryotes), where undulations, you might ask whether this esoteric example they are responsible for locomotion. Their behavior is is of any interest to the neuroscience community, since, as similar to those in our lungs, except that the flow of fluid mentioned above, such extreme reliance on sensory feed- in one direction propels the protozoa in the other. back and absence of CPGs is truly unprecedented in the One of the most exciting applications of the theory animal world. However, it is important to remember that of undulatory propulsion in recent years is to the field

14 ContemporaryPhysics,toappear 7 DISCUSSION of bio-inspired robotics. Traditionally robots have used different simulation approaches, to biological examples wheels or tracks to get around, but neither of these of undulating organisms. The field is too vast to have are a great solution to the problem of locomotion in touched on all the research that is relevant and ongoing, heterogeneous and difficult terrain. Given the success so we have chosen to offer only a cursory view of better with which biological organisms move around, it is known systems and results (such as bacterial swimming natural to attempt to emulate these behaviors. In many and lamprey CPGs), and to spend more time on less fa- cases, legged locomotion seems like the ideal solution miliar examples, such as Helmholtz’s minimum energy and recent advances suggest that legged robots may soon dissipation theorem, generalizations of the scallop theo- be sufficiently advanced to handle real terrain. But there rem and nematode locomotion. are also many cases where undulatory locomotion would be preferable. A snake-like robot89 may, for example, To the physicist uninitiated in biological parlance, the be able to inspect the inside of narrow pipes, search for level of biological detail in some of the examples may ap- survivors among the rubble of collapsed buildings and pear daunting, or perhaps superfluous. However, one of traverse even the roughest terrain. A snail-like robot the key roles of the biological physicist is to tease out (snail locomotion uses a form of undulation)90 could the essential features of what can seem an almost arbi- climb vertically or even upside down. But a particularly trarily complicated system. Therefore, we have not shied interesting environment in which small, worm-like away from leaving the physics grounded in at least some robots91 could potentially operate is the human body92. of the biological systems in this review. We hope that Designers of medical technologies constantly strive to this approach serves less as a hindrance and more as an find non-invasive (or less invasive) ways to diagnose and invitation to the readers to engage in the larger endeavor repair problems within our bodies, and micro-robots that is the interface between physics and biology. (or even remotely guided bacteria93) are a promising avenue. Undulatory locomotion94 is ideally suited to use within the human body as it is generally robust to heterogeneous environments and is non-destructive. While we are still a long way from this final goal, researchers working on small scale undulatory robots are following a number of interesting directions and making strides both on foundational issues and on the Acknowledgments non-trivial engineering challenges of miniaturization.

7. DISCUSSION The authors are grateful for useful discussions with Robert McNeill Alexander, Duncan Dowson, Sam Braun- In this paper, we have gone from reviewing some fun- stein and Stefano Berri. NC acknowledges support from damental physics applied to low Re swimmers, through the EPSRC and BBSRC.

∗ Electronic address: [email protected] 13 R. D. Maladen, Y. Ding, C. Li, and D. I. Goldman, Science † Electronic address: [email protected] 325, 314 (2009). 1 W. Baumgartner, F. Fidler, A. Weth, M. Habbecke, 14 W. Ludwig, Zeitschrift f¨ur Vergleichende Physiologie 13, P. Jakob, C. Butenweg, and B¨ohme, PLoS ONE 3, e3309. 397 (1930). doi:10.1371/journal.pone.0003309 (2008). 15 E. M. Purcell, Am. J. Phys. 45, 3 (1977). 2 H. C. Berg, Physics Today 53, 24 (2000). 16 A. Shapere and F. Wilczek, J. Fl. Mech. 198, 557 (1989). 3 H. C. Berg, E. coli in Motion (Springer, New York, 2004), 17 J. E. Avron, O. Kenneth, and D. H. Oaknin, New J. Phys. 1st ed. 7, 234 (2005). 4 H. A. Stone and A. D. T. Samuel, Am. J. Phys. 45, 3 18 G. I. Taylor, Proc. Roy. Soc. A 209, 447 (1951). (1996). 19 G. I. Taylor, Proc. Roy. Soc. A 214, 158 (1952). 5 K. M. Ehlers, A. D. T. Samuel, H. C. Berg, and R. Mont- 20 G. J. Hancock, Proc. Roy. Soc. A 217, 96 (1953). gomery, Proc. Natl. Acad. Sci. U.S.A. 93, 8340 (1996). 21 J. Gray, Quart. J. Micro. Sci. 94, 551 (1953). 6 C. M. Breder, Zoologica 4, 159 (1926). 22 H. R. Wallace, Ann. Appl. Biol. 46, 74 (1958). 7 J. J. Head, J. I. Bloch, A. K. Hastings, J. R. Bourque, E. A. 23 H. R. Wallace, Ann. Appl. Biol. 47, 366 (1959). Cadena, F. A. Herrera, P. D. Polly, and C. A. Jaramillo, 24 H. R. Wallace, Ann. Appl. Biol. 48, 107 (1960). Nature 457, 715 (2009). 25 J. Lighthill, Proc. R. Soc. Lond. B. 179, 125 (1971). 8 G. La Spina, M. Sfakiotakis, D. P. Tsakiris, A. Menciassi, 26 R. G. Cox, J. Fluid Mec. 44, 791 (1970). and P. Dario, IEEE Trans. Robot. 23, 1200 (2007). 27 J. Lighthill, SIAM Rev. 18, 161 (1976). 9 T. L. Jahn, M. D. Landman, and J. R. Fonseca, J. Proto- 28 J. Gray and G. J. Hancock, J. Exp. Biol. 32, 802 (1955). zool. 11, 291 (1964). 29 J. Gray and H. W. Lissmann, J. Exp. Biol. 41, 135 (1964). 10 U. M¨uller, Science 302, 1511 (2003). 30 R. E. Johnson and C. J. Brokaw, Biophys. J. 25, 113 11 J. C. Liao, D. N. Beal, G. V. Lauder, and M. S. Triantafyl- (1979). lou, Science 302, 1566 (2003). 31 S. Berri, J. H. Boyle, M. Tassieri, I. A. Hope, and N. Co- 12 J. C. Liao, Phil. Trans. Roy. Soc. B - Biol. Sci. 362, 1973 hen, HFSP J. 3, 186 (2009). (2007). 32 A. Looss, Cairo Records Eg. Govt. Sch. Med. 4, 167 (1911).

15 ContemporaryPhysics,toappear 7 DISCUSSION

33 A. M. Horner and B. C. Jayne, J. Exp. Biol. 211, 1612 Do nmoyer, and C. D. Johnson, Trends Neurosci. 8, 294 (2008). (1985). 34 S. R. Lockery, K. J. Lawton, J. C. Doll, S. Faumont, S. M. 70 A. J. Ijspeert, J. Hallam, and D. Willshaw, Adapt. Behav. Coulthard, N. Thiele, N. Chronis, K. E. McCormick, M. B. 7, 151 (1999). Goodman, and B. L. Pruitt, J. Neurophysiol. 99, 3136 71 A. H. Cohen, G. B. Ermentrout, T. Kiemel, N. Kopell, (2008). K. A. Sigvardt, and T. L. Williams, Trends Neurosci. 15, 35 J. Pierce-Shimomura, B. L. Chen, J. J. Mun, R. Ho, 434 (1992). R. Sarkis, and S. L. McIntire, Proc. Natl. Acad. Sci. U.S.A. 72 A. J. Ijspeert, A. Crespi, and J. M. Cabelguen, Neuroin- 105, 20982???20987 (2008). formatics 3, 171 (2005). 36 D. G. Morgan and S. Khan, Bacterial Flagella (John Wiley 73 N. Kopell and G. B. Ermentrout, Commun. Pur. Apply. & Sons, Chichester, 2001), p. http://www.els.net/ DOI: Math. 39, 623 (1986). 10.1038/npg.els.0000301. 74 O. Ekeberg, Biol. Cyb. 69, 363 (1993). 37 M. L. Ginger, N. Portman, and P. G. McKean, Nat. Rev. 75 A. Crespi, A. Badertscher, A. Guignard, and A. J. Ijspeert, Microbiol. 6, 838 (2008). Robot. Auton. Syst. 50, 163 (2005). 38 Y. Sowa and R. M. Berry, Q. Rev. Biophys. 41, 103 (2008). 76 A. H. Cohen and D. Boothe, Auton. Robot. 7, 239 (1999). 39 D. Eshel, C. Shingyoji, K. Yoshimura, B. H. Gibbons, I. R. 77 M. F. Simoni and S. P. DeWeerth, IEEE Trans. Biomed. Gibbons, and K. Takahashi, J. Exp. Biol. 152, 441 (1990). Eng. 54, 193 (2007). 40 A. G. Loewy, P. Siekevitz, J. R. Menninger, and J. A. N. 78 J. H. Boyle, S. Berri, I. A. Hope, and N. Cohen, p. Unpub- Gallan, Cell Structure and Function (Saunders College lished (2009). Publishing, Orlando, Florida, 1991), 3rd ed. 79 J. A. Bryden and N. Cohen, Biol. Cyb. 98, 339 (2008). 41 G. I. Taylor, Low-Reynolds number flows, 6mm colour 80 J. Karbowski, G. Schindelman, C. Cronin, A. Seah, and sound film, produced by Educational Serviced, Inc. (1967). P. Sternberg, J. Comp. Neurosci. 24, 253 (2008). 42 H. C. Berg, Proc. Natl. Acad. Sci. U.S.A. 93, 14225 (1996). 81 W. R. Schneider and R. N. Doetsch, J. Bacteriol. 117, 696 43 R. M. MacNab and M. K. Ornston, J. Mol. Biol. 112, 1 (1974). (1977). 82 C. J. Brokaw, J. Exp. Biol. 45, 113 (1966). 44 T. Vicsek, Fluctuations and Scaling in Biology (Oxford 83 D. M. Wilson, J. Exp. Biol. 38, 471 (1961). University Press, 2001), 1st ed. 84 K. G. Pearson, Curr. Opin. Neurobiol. 5, 786 (1995). 45 H. R. Wallace, Annu. Rev. Phytopathol. 6, 91 (1968). 85 X. YU, B. Nguyen, and W. O. Friesen, J. Neurosci. 19, 46 E. Niebur and P. Erd¨os, Biophys. J. 60, 1132 (1991). 4634 (1999). 47 H. R. Wallace, Nematologica 15, 65 (1969). 86 J. D. Huizinga and W. J. E. P. Lammers, 296, G1 (2009). 48 G. D. Kirchhoff, Ann. Phys. 75, 189 (1848). 87 D. Bray, Cell Movements: from molecules to motility (Tay- 49 H. N. Helmholtz, Crelle’s J. 57, 1132 (1859). lor and Francis, London, 2001), 2nd ed. 50 H. N. Helmholtz, Verhandlungen des naturhistorisch- 88 M. R. Marino and E. Aiello, Cell Motility Supplement 1, medicinischen Vereins zu Heidelberg (1868). 35 (1982). 51 D. J. Korteweg, Phil. Mag. 16, 112 (1883). 89 J. K. Hopkins, B. W. Spranklin, and S. K. Gupta, Bioinsp. 52 G. B. Jeffrey, Proc. Roy. Soc. A 102, 161 (1922). Biomim. 4, 1 (2009). 53 G. I. Taylor, Proc. Roy. Soc. A 103, 61 (1923). 90 B. Chan, N. J. Balmforth, and A. E. Hosoi, Phys. Fluids 54 D. G. Christopherson and H. Naylor, Proc. Inst. Mech. 17, 113101 (2005). Engrs. 169 (1955). 91 A. Menciassi, D. Accoto, S. Gorini, and P. Dario, Auton. 55 D. Dowson and D. G. Christopherson, Proc. Roy. Soc. A Robot. 21, 155 (2006). 251 (1959). 92 B. Chen, Y. Liu, S. Chen, S. Jiang, and H. Wu, J. Bion. 56 J. White, E. Southgate, J. Thomson, and S. Brenner, Phi- Eng. Suppl. 5, 106 (2008). los. Trans. Roy. Soc. London B Biol. Sci. 314, 1 (1986). 93 S. Martel, C. C. Tremblay, S. Ngageng, and G. Langlois, 57 B. Chen, D. Hall, and D. Chklovskii, Proc. Natl. Acad. Appl. Phys. Lett. 89, 233904 (2006). Sci. U.S.A. 103, 4723 (2006). 94 B. Behkam and M. Sitti, J. Dyn. Syst. T. ASME 128, 36 58 E. Marder and R. L. Calabrese, Physiol. Rev. 76, 687 (2006). (1996). 95 In addition to being linear, Eq. (1) stipulates that fluid 59 E. Marder, Curr. Opin. Neurobiol. 10, 691 (2000). that is infinitessimally close to the spoon will travel at the 60 R. M. Harris-Warrick, Curr. Opin. Neurobiol. 3, 982 same speed as the spoon. This is often called the no-slip (1993). boundary condition (see also Sec. 2.3). 61 A. H. Cohen and P. Wallen, Exp. Brain Res. 41, 11 (1980). 96 Strictly speaking the Reynolds number is defined as the ra- 62 A. H. Cohen, P. J. Holmes, and R. H. Rand, J. Math. Biol. tio of the viscous to the convective (nonlinear) term, but in 13, 345 (1982). the low Re regime, both inertial terms can be neglected15. 63 A. H. Cohen, J. Neurosci. Meth. 21, 113 (1987). 97 This is exactly analogous to the solution of the time- 64 N. Kopell and G. B. Ermentrout, Math. Biosci. 90, 87 dependent Schr¨odinger equation in terms of the time or- (1988). dered exponential. 65 G. B. Ermentrout and N. Kopell, SIAM J. Appl. Math. 98 Note that visco-elasticity is typically modeled by adding 54, 478 (1994). time-dependent terms to the Navier-Stokes equation, so 66 S. Grillner, D. Parker, and A. el Manira, Ann. NY. Acad. this parametrization by K is only valid in the limit where Sci. 860, 1 (1998). this time dependence can be neglected. 67 A. H. Cohen, J. Comp. Physiol. A 160, 181 (1987). 99 Indeed, worms moving in highly resistive environments are 68 P. Churchland and T. J. Sejnowski, The Computational observed to have higher average curvature, shorter wave- Brain (MIT Press, Cambridge, 1992). lengths and lower amplitudes than in water. 69 A. O. W. Stretton, R. E. Davis, J. D. Angstadt, J. E.

16 ContemporaryPhysics,toappear