PHYSICAL REVIEW E 80, 021903 共2009兲 Noisy swimming at low Reynolds numbers Jörn Dunkel* and Irwin M. Zaid Rudolf Peierls Centre for Theoretical Physics, University of Oxford, 1 Keble Road, Oxford OX1 3NP, United Kingdom 共Received 30 March 2009; published 5 August 2009兲 Small organisms 共e.g., bacteria兲 and artificial microswimmers move due to a combination of active swimming and passive Brownian motion. Considering a simplified linear three-sphere swimmer, we study how the swimmer size regulates the interplay between self-driven and diffusive behavior at low Reynolds number. Starting from the Kirkwood-Smoluchowski equation and its corresponding Langevin equation, we derive formulas for the orientation correlation time, the mean velocity and the mean-square displacement in three space dimensions. The validity of the analytical results is illustrated through numerical simulations. Tuning the swimmer parameters to values that are typical of bacteria, we find three characteristic regimes: 共i兲 Brownian motion at small times, 共ii兲 quasiballistic behavior at intermediate time scales, and 共iii兲 quasidiffusive behavior at large times due to noise-induced rotation. Our analytical results can be useful for a better quantitative understanding of optimal foraging strategies in bacterial systems, and they can help to construct more efficient artificial microswimmers in fluctuating fluids. DOI: 10.1103/PhysRevE.80.021903 PACS number共s兲: 87.17.Jj, 05.40.Jc, 47.63.Gd, 47.63.mf I. INTRODUCTION Biological 关1–3兴 and artificial microswimmers 关1,4–7兴 move through a fluid by performing a series of self-induced shape changes 关8,9兴. Handicapped by their tiny size 共typically a few micrometers for a bacterium 关10兴兲, they are forced to swim at very low Reynolds numbers R Ⰶ 1 关11–13兴. Hence, in order to account for the resulting lack of inertia, the swimming strategies of microorganisms are very different from those operative at human length scales. More precisely, since the fluid flow is reversible at low Reynolds number, locomotion in this regime is possible only if the swimming stroke violates certain time-reversal symmetries 关8,13–16兴. Stimulated by experimental advances 关1–4,17,18兴, in recent years considerable progress was achieved in understanding the dynamics of deterministic microswimmers models 关19–28兴. Yet, comparatively little is known quantitatively about the complex interplay between active self-motion, hydrodynamic interactions, and thermal fluctuations in the surrounding fluid 关29–32兴. Very recently, first steps toward clarifying these issues were made by Howse et al. 关6兴, who measured in their experiments the mean square displacement of chemically driven colloidal spheres, and by Lobaskin et al. 关33兴, who studied the Brownian motion of a triangular microswimmer at intermediate Reynolds numbers R ⬃ 1 by combining Lattice Boltzmann simulations with a Langevin description of the swimmer in phase space. In the present paper, we would like to complement these investigations by concentrating on the diffusive properties of mechanically driven microswimmers at low Reynolds numbers R Ⰶ 1. This limit case is most relevant for bacterial motions and allows one to treat diffusive effects within configuration space. Specifically, we will focus on the following questions: how does the size of the swimmer affect its effective mobil- *jorn.dunkel@physics.ox.ac.uk 1539-3755/2009/80共2兲/021903共11兲 ity in a noisy fluid? Which details govern the transition from quasiballistic self-motion to the diffusive regime? To shed light on these issues, we shall consider simplified quasilinear p-sphere swimmers similar to those proposed by Najafi and Golestanian 关20兴. More precisely, we will assume that internal forces, which generate the swimming strokes, are mediated by interaction potentials. This approach permits us to treat thermal diffusion effects within the KirkwoodSmoluchowski scheme, originally developed to describe the diffusion of polymers in a fluctuating medium 关34–37兴. Starting from the Kirkwood-Smoluchowski equation 共KSE兲 ensures that hydrodynamic and stochastic forces are consistently coupled on the level of the Fokker-Planck description in configuration space 关34兴. Moreover, as discussed in Sec. III, the corresponding Langevin equation can be used to derive closed analytical formulas for the orientation correlation time, the mean velocity and the mean square displacement of a single swimmer in three space dimensions. Although the analytical and numerical results in this paper refer to the case of a quasilinear three-sphere swimmer 共p = 3兲, the formalism can be easily generalized to more complex models 共e.g., flexible p-sphere swimmer chains兲. Therefore, this approach can be generally very useful for studying Brownian motion effects and hydrodynamic interactions in active biological systems at low Reynolds numbers. Furthermore, since it is straightforward to implement an external confinement 共e.g., tweezer or lattice potentials兲, the Kirkwood-Smoluchowski scheme can help to construct and optimize arrays 关38兴 of, e.g., micropumps that work efficiently on those scales where thermal fluctuations in fluid become non-neglible. Thus, purpose and content of the present paper can be summarized as follows: first, we will discuss a convenient formalism that allows to simulate active microswimming by means of Langevin equations and interaction potentials 共Sec. II兲. Subsequently, we derive analytic results for the diffusion of a single swimmer 共Sec. III兲, thereby extending recent work of Golestanian and Ajdari 关25兴 on deterministic swimmers. A thorough analysis of the single swimmer case is instructive for a number of reasons: 共i兲 Exact analytical re- 021903-1 ©2009 The American Physical Society PHYSICAL REVIEW E 80, 021903 共2009兲 JÖRN DUNKEL AND IRWIN M. ZAID sults provide a useful test of numerical simulations, cf. Sec. IV. 共ii兲 Recently, Leoni et al. 关4兴 were able to experimentally realize an individual three-sphere system similar to those considered here. 共iii兲 Understanding the noise-induced behavior of a single swimmer is a prerequisite for understanding complex behavior and pattern formation in, e.g., selfassembling bacterial systems 关10兴. 共iv兲 Dynamical calculations as those presented below may provide a “microscopic” justification for purely probabilistic models of bacterial motility 关39兴. 共v兲 Depending on the swimmer size, we find a rather sharp transition from purely Brownian to quasiballistic motions. From a 共bio兲 physical perspective, it is remarkable that the transition occurs when the three-sphere swimmer model is tuned to bacterial parameters. Hence, loosely speaking, exploiting the interplay between Brownian motion and active swimming may indeed represent a useful strategy 共not only兲 in nature. II. GENERAL THEORETICAL BACKGROUND We consider an ensemble of N microswimmers, each consisting of p spheres. Neglecting inertia, the state of the system at time t is described by a set of coordinates 共X␣兲 = 兵X共␣i兲共t兲其, where ␣ = 1 , . . . , pN is a sphere index, and i = 1 , 2 , 3 labels the space dimension. Our subsequent analysis rests on the assumption that the stochastic dynamics of the swimmers in the fluid can be described, at least approximately, by the KSE 关34–37兴, representing evolution for the N-particle probability density f共t , 兵x共␣i兲其兲. We begin by recalling how the KSE can be translated into a Langevin equation for numerical simulations 关36,37兴. Details of the swimming mechanism will be discussed in Sec. II B. Here, we are interested in the effects of hydrodynamic interactions, corresponding to H共␣i兲共 j兲 ⫽ 0. A simple approximation for H共␣i兲共 j兲, obtained by solving the Stokes equation for a pointlike source, is the Oseen tensor 关40,41兴 H共O␣i兲共 j兲 = 冉 冊 1 r r ␦ij + ␣2i ␣ j , 8r␣ r␣ ␣ ⫽ , 共2b兲 where r␣i ª x␣i − xi and r␣ ª 兩x␣ − x兩. However, the associated diffusion tensor DO ª kBTHO is not necessarily positive definite, leading to unphysical behavior if sphere separations become too small 关42,43兴. In our numerical simulations we shall therefore use the improved approximation H共M␣i兲共 j兲 = H共O␣i兲共 j兲 + 共a␣2 + a2 兲 3 24r␣ 冉 ␦ij − 3 冊 r␣ir␣ j , 共2c兲 2 r␣ which was derived by Mazur 关44兴. The additional term on the right-hand side of Eq. 共2c兲 can be understood as the next-order correction in a radius-over-distance expansion of the mobility tensor for two spheres 关45兴. For spheres of equal size 共a␣ = a兲, the tensor H = HM defined by Eqs. 共2a兲 and 共2c兲, reduces to the Rotne-Prager-Yamakawa tensor 关43,46兴. While Eq. 共2c兲 gives a more accurate description than Eq. 共2b兲 at moderate densities, both expressions become invalid if the distance between spheres becomes very small. At very high densities, when sphere-sphere collisions dominate the dynamics, near-field hydrodynamics and lubrication effects must be modeled more carefully 关47兴. In the present paper, however, we focus on systems that can be described by Eq. 共2c兲. Unlike the Oseen tensor HO from Eq. 共2b兲, the tensor H = HM is positive definite for r␣ ⬎ a␣ + a and thus can be Cholesky decomposed in the form A. Kirkwood-Smoluchowski and Langevin equation Considering a fluid of viscosity and temperature T, the KSE reads 关34兴 1 H共␣i兲共 j兲 = C共␣i兲共␥k兲C共 j兲共␥k兲 . 2 t f = 共␣i兲H共␣i兲共 j兲兵关共 j兲U兴f + kBT共 j兲 f其. The decomposition Eq. 共3兲 is crucial if one wishes to find a Langevin representation for the stochastic process 兵X共␣i兲共t兲其 described by the KSE Eq. 共1兲: upon noting that 关36,48兴 共1兲 Here, kB denotes the Boltzmann constant, 共␣i兲 ª / x共␣i兲, and a sum is performed over equal double indices 共␣i兲. The 共effective 关34兴兲 potential U governs all the internal and external swimmer interactions 共see examples in Sec. II B兲, apart from the hydrodynamic interactions mediated by the fluid. The latter are included in the tensor H. Considering spherical particles of radius a␣, the “diagonal” components of H are given by ␦ij H共␣i兲共␣ j兲 = , ␥␣ ␥␣ = 6a␣ , 共2a兲 where ␦ij denotes the Kronecker symbol, and ␥␣ is the Stokes friction coefficient. The hydrodynamic interactions between different spheres are encoded in the “off-diagonal” components H共␣i兲共 j兲, ␣ ⫽ . If these hydrodynamic interactions are neglected, corresponding to the 共infinitely dilute兲 limit case H共␣i兲共 j兲 = 0, Eq. 共1兲 reduces to an “ordinary” Smoluchowski equation with a diffusion constant D␣ = kBT / ␥␣ for each sphere. 共␣i兲H共␣i兲共 j兲 ⬅ 0 共3兲 共4兲 holds 共for both HO and HM 兲, one finds that the KSE Eq. 共1兲 corresponds to the following Ito-Langevin equation 关49兴 dX共␣i兲共t兲 = H共␣i兲共 j兲F共 j兲dt + 共kBT兲1/2C共␣i兲共␥k兲dB共␥k兲共t兲. 共5兲 Here, 兵F共 j兲其 ª 兵−共 j兲U其 comprises the deterministic forces acting on the spheres, and 兵B共␥k兲共t兲其 is a collection of standard Wiener processes; i.e., the increments dB共␥k兲共t兲 ª B␥k 共t + dt兲 − B共␥k兲共t兲 are independent GAUSSIAN random numbers with distribution 2 P兵dB共␥k兲共t兲 苸 关u,u + du兴其 = e−u /共2dt兲 du, 共2dt兲1/2 共6兲 and, according to the Ito scheme 关49兴, the coefficients C共␣i兲共␥k兲 are to be evaluated at time t. For completeness, we still note that, upon formally dividing by dt, the stochastic 021903-2 PHYSICAL REVIEW E 80, 021903 共2009兲 NOISY SWIMMING AT LOW REYNOLDS NUMBERS differential Eq. 共5兲 can be rewritten as a “standard” Langevin equation Ẋ共␣i兲共t兲 = H共␣i兲共 j兲F共 j兲 + 共kBT兲1/2C共␣i兲共␥k兲共␥k兲共t兲, 共7a兲 where Ẋ共␣i兲共t兲 ª dX共␣i兲共t兲 / dt denotes the velocity, and 共␥k兲共t兲 ª dB共␥k兲共t兲 / dt represents GAUSSIAN white noise, i.e., 具共␣i兲共t兲典 = 0, 共7b兲 具共␣i兲共t兲共 j兲共t⬘兲典 = ␦␣␦ij␦共t − t⬘兲. 共7c兲 B. Swimming mechanism Having discussed how to implement fluctuations, we still need to specify the swimming mechanism. To this end, consider two spheres ␣ and  that form the leg of a swimmer. We assume that the internal forces between beads ␣ and , which generate the swimming stroke, can be derived from a time-dependent potential of the form Uleg共t,d␣兲 = k0 兵d␣ − 关ᐉ + ␣ sin共t + ␣兲兴其2 , 共8a兲 2 with d␣共t兲 ª 兩X␣共t兲 − X共t兲兩 denoting the distance between the spheres, ␣ the approximate amplitude of the stroke, and ᐉ ⬎ a␣ + a + ␣ the mean length of the leg. The potential Uleg gives rise to two characteristic time scales: the driving period T ª 2 / and, for a sphere of mass M ␣, the oscillator period T0 ª 2 / 冑k0 / M ␣. Since we are interested in the overdamped regime, these time scales must be long compared to the characteristic damping time T␥ ª M ␣ / ␥␣. More precisely, we have to impose T␥ Ⰶ T0 Ⰶ T n␥ ª d␥/d␥ . In order to ensure that the swimmer moves quasilinearly 关25兴, we introduce a stiffness potential K 2 ᐉ 共n␣ · n␥ − 1兲2 , 2 Q 共␦ik − n␥in␥k兲n␣i , d␥ Flink = − 共F␣link + F␥link兲 共10b兲 共10c兲 where n␣i ª d␣i / d␣, and Q ª Kᐉ2共n␣ · n␥ − 1兲. 共10d兲 Equations 共8兲 and 共9兲 provide a convenient way of modeling and simulating rigid or flexible p-sphere swimmers by means of potentials. We note that, by construction, the sum over the internal swimmer forces is zero. The total potential U appearing in the KSE Eq. 共1兲 is obtained by summing over all effective interaction potentials Eqs. 共8兲 and 共9兲. III. ANALYTICAL RESULTS We next summarize formulae for the correlation time of the orientation vector, the mean swimmer velocity and the spatial mean square displacement of an isolated three-sphere swimmer. These analytical results can be obtained from the Langevin Eq. 共5兲 by using the Oseen approximation H ⯝ HO, and their explicit derivation is discussed in the Appendix. A swimmer’s motion can be characterized by its geometric center 1 R共t兲 ª 共X1 + X2 + X3兲 3 共11a兲 and the orientation vector N共t兲 ª 共8b兲 corresponding to slow driving and fast relaxation. The constraint Eq. 共8b兲 ensures that our swimmers behave similar to a shape-driven swimmer 关25兴. In the remainder, we shall focus on three-sphere swimmers, representing the smallest self-swimming system within our approach 共two-sphere swimmers can achieve active locomotion only due to collective effects 关16兴兲. We consider three spheres 共␣ ,  , ␥兲 forming a swimmer with central sphere , e.g., in the case of a single swimmer 共␣ ,  , ␥兲 = 共1 , 2 , 3兲 with middle sphere  = 2. The legs are given by d␣ ª X − X␣ and d␥ ª X␥ − X, and we still define normalized connectors n␣ ª d␣/d␣, F␥link = X3 − X1 . 兩X3 − X1兩 共11b兲 We are interested in determining the mean square displacement DR共t兲 ª 具关R共t兲 − R共0兲兴2典, 共12a兲 and the correlation function DN共t兲 ª 具N共t兲N共0兲典, 共12b兲 where the average is taken over fluctuations in the fluid 共i.e., over all realizations of the Wiener process兲. It is convenient to discuss DN共t兲 first. For a deterministic initial state N共0兲 = 关Nk共0兲兴, we can write DN共t兲 = 具Nk共t兲典Nk共0兲 with a summation over equal indices. To obtain an analytical formula for 具Nk共t兲典, we assume that Eq. 共8b兲 holds true and that bending is neglible, K Ⰷ k0. Then the swimmer behaves like a stiff, shape-driven NajafiGolestanian 关20兴 swimmer and we can approximate 共9兲 d12 ª X2 − X1 ⯝ Nd12 , 共13a兲 which, for K Ⰷ k0, penalizes bending. The resulting force components F共lin␣i兲 ª −共␣i兲Ulin read explicitly d23 ª X3 − X2 ⯝ Nd23 , 共13b兲 d13 ª X3 − X1 ⯝ N共d12 + d23兲, 共13c兲 Ulin = F␣link = − Q 共␦ik − n␣in␣k兲n␥i , d␣ 共10a兲 where, cf. Eq. 共8a兲, 021903-3 PHYSICAL REVIEW E 80, 021903 共2009兲 JÖRN DUNKEL AND IRWIN M. ZAID d12 = ᐉ + 12 sin共t + 12兲, 共13d兲 d23 = ᐉ + 23 sin共t + 23兲. 共13e兲 Adopting the Oseen approximation H ⯝ H , one can derive from the Langevin Eq. 共5兲 the following linear evolution equation 共see App. A1兲 B3 ª O 具Ṅk典 = − k BT 3 2d13 冋冉 冊 册 2 d13 d13 + − 1 具Nk典, 3 a1 a3 共14兲 where 具Ṅk典 ª 具dNk共t兲 / dt典. The 1 / a␣-parts are contributions to the rotation rate due to noise on the spheres, whereas the 3 contribution is a correction due to hydrodynamic inter1 / d13 actions. Equation 共14兲 can be solved exactly. The solution exhibits an exponentially decaying oscillatory behavior due to the periodicity of the swimming stroke d13. However, for ᐉ Ⰷ max兵␣其 it suffices to approximate d13 ⯝ 2ᐉ, yielding an exponential decay DN共t兲 ⯝ exp共−t / N兲 with orientation correlation time N ⯝ 再 冋 冉 冊 册冎 k BT 16ᐉ3 4 ᐉ ᐉ + −1 3 a1 a3 . 共15兲 共16a兲 where B3具ḋ12k典 + C具ḋ23k典 , B 1B 3 − C 2 共16b兲 共16c兲 are the noise-averaged internal forces on the first and third sphere 共the force on the central sphere can be eliminated by virtue of F共1k兲 + F共2k兲 + F共3k兲 ⬅ 0兲, respectively, and 具ḋ12k典 = 具Ṅk典d12 + 具Nk典ḋ12 , 共16d兲 具ḋ23k典 = 具Ṅk典d23 + 具Nk典ḋ23 共16e兲 the mean change in the leg vectors due stochastic rotations and swimming strokes, with abbreviations 冊 冊 1 1 1 1 1 A1 ª − + − , ␥1 ␥2 4 d13 d23 1 1 1 1 1 − + − , A3 ª ␥3 ␥2 4 d13 d12 1 1 1 + − , ␥1 ␥2 2d12 共16j兲 具Ṙ共t兲典 = V具N共t兲典. 共17a兲 Here, V denotes the stroke-averaged velocity 共i.e, over an interval 关t , t + T兴兲 of the corresponding deterministic swimmer 关25兴. By using the approximation Eq. 共17a兲 instead of the exact results Eq. 共16兲 one neglects mean velocity oscillations on small time scales. For example, when considering equal-sized beads with a␣ = a and ᐉ Ⰷ max兵a , ␣其, then 冉 冊 7 1223 sin ⌬ , a 24 ᐉ2 共17b兲 where ⌬ ª 12 − 23 is the phase difference of the leg contractions, and higher-order terms in a and ␣ are neglected. Integrating Eq. 共17兲 with 具N共t兲典 ⯝ N共0兲exp共−t / N兲, we obtain for the position mean value of the swimmer the simple approximate result 具R共t兲典 ⯝ R共0兲 + VN关1 − exp共− t/N兲兴N共0兲; C具ḋ12k典 + B1具ḋ23k典 具F共3k兲典 = , B 1B 3 − C 2 B1 ª 冊 Since the quantities 具Nk典, 具Ṅk典, d␣, and ḋ␣ are known, Eqs. 共16兲 provide a closed analytical result for the mean velocity 具Ṙk典 of a shape-driven swimmer 共within the Oseen approximation兲. In particular, Eqs. 共16兲 generalize the corresponding velocity formulas for a deterministic swimmer, recently obtained by Golestanian and Ajdari 关25兴, to the “noisy swimming” regime. For realistic swimmer parameters the orientation correlation time N is typically much larger than the driving period T = 2 / . In this case, the rather lengthy result Eq. 共16兲 can be considerably simplified 共see last part of App. A2a for details兲 to read V= A1 A3 具Ṙk典 = 具F共1k兲典 + 具F共3k兲典, 3 3 冉 冉 冉 共16i兲 1 1 1 1 1 − + − . ␥2 4 d12 d23 d13 −1 The time parameter N not only determines the temporal correlation of the orientation vector, it also plays an important role for the dynamics of the geometric center R共t兲. As shown in Appendix, the mean swimmer velocity 具Ṙ共t兲典 is governed by the equation 具F共1k兲典 = − Cª 1 1 1 + − , ␥2 ␥3 2d23 i.e., in the asymptotic limit t → ⬁, 具R共⬁兲典 = R共0兲 + VNN共0兲. 共19兲 Finally, let us still consider the mean-square displacement DR共t兲 ª 具关R共t兲 − R共0兲兴2典 for a stiff, shape-driven three-sphere swimmer described by Eqs. 共13兲. As discussed in App. A2b, by starting from the Langevin equation for R共t兲, one can show that DR共t兲 decomposes into the form DR共t兲 = DRp 共t兲 + DRa 共t兲, where the first part DRp 共t兲 = 共16f兲 共16g兲 冉 冊 冊 1 k BT 1 1 1 2 k BT + + t+ 9 a1 a2 a3 9 ⫻ds 共16h兲 共18兲 冉 1 1 1 + + d12 d23 d13 共20a兲 冕 t 0 共20b兲 comprises passive Brownian motion contributions due to thermal diffusion of the spheres 共first line兲 and hydrodynamic Oseen interactions between them 共second line兲, while the second part 021903-4 PHYSICAL REVIEW E 80, 021903 共2009兲 NOISY SWIMMING AT LOW REYNOLDS NUMBERS 冕 冕 t DRa 共t兲 ⯝ V2 t ds du具N共s兲N共u兲典 共20c兲 0 0 is the contribution due to active self-swimming. Similar to Eq. 共17兲, the expression Eq. 共20c兲 is valid if the orientation correlation time is much larger than the stroke period, N Ⰷ T. Inserting the above result 具N共t兲N共s兲典 ⯝ exp共−t / N兲 and approximating d12 = d23 = d13 / 2 ⯝ ᐉ, we obtain for the spatial mean-square displacement DR共t兲 ⯝ 冉 冊 1 k BT 1 1 1 5 + + + t 9 a1 a2 a3 ᐉ + 2V2N关t + N共e−t/N − 1兲兴. 共21兲 The approximate result Eq. 共21兲 provides a coarse-grained stroke-averaged description of the translational diffusion 共details of the swimming stroke are encoded in V兲. An analogous formula can be used to describe the diffusion of a spherical, chemically driven microswimmer 关6兴. By virtue of Eq. 共21兲, we can readily distinguish three distinct regimes: 共i兲 For t Ⰶ N we can expand the exponential term to linear order and find DR共t兲 ⯝ DRp 共t兲, IV. SWIMMER TUNING & NUMERICAL SIMULATIONS We simulate a single three-sphere swimmer described by the interaction potentials Eqs. 共8兲 and 共9兲 and governed by the Langevin Eq. 共5兲. We consider identical spheres of radius a␣ = a, mass M ␣ = M, and equal stroke amplitudes 12 = 23 = . The density of the spheres is chosen as = 103 kg/ m3 共water兲, and the fluid is water at room temperature 关 = 10−3 kg/ 共ms兲 , T = 300 K兴. By fixing the parameters of the spring and bending potentials as k0 = 10−4 kg/ s2, = 103 Hz, and K = 10k0, we satisfy the time scale condition Eq. 共8b兲 while ensuring that the swimmer behaves approximately stiff and shape-driven. The velocity V of the swimmer is optimized by choosing ⌬ = / 2, cf. Eq. 共23a兲. We are primarily interested in understanding how a change in the swimmer size may affect the diffusive behavior. We therefore fix the ratios ᐉ / a = ᐉ / = 10 and only vary the leg length ᐉ from 1 to 10 m in our simulations 共i.e., the mean swimmer length is ⌳ = 2ᐉ兲. Put differently, we scale the swimmer proportionally by varying ᐉ. Having specified all parameters, it is useful to summarize the relevant formulae for our choice V⯝ 共22a兲 i.e., passive Brownian motion dominates on very short time scales. 共ii兲 For t ⱗ N we need to include terms quadratic in t / N and obtain Vt N ⯝ DRp t 2 2 DR共t兲 ⯝ DRp 共t兲 + N , 共22b兲 i.e., ballistic motion can dominate on intermediate time scales provided V2 / N is large enough 共cf. examples below兲. 共iii兲 For t Ⰷ N, we recover diffusive behavior lim t→⬁ DR共t兲 t = DRp 共t兲 t + 2V2N . 共22c兲 If the swimmer is constructed such that 2V2N Ⰷ DRp 共t兲 / t, then its diffusive behavior at large times is due to noiseinduced rotation 共with persistence time N兲. In a sense, the above results provide a “microscopic” justification for the assumptions made by Lovely and Dahlquist 关39兴, who studied purely probabilistic models of bacterial motion. We also note that the asymptotic behavior for t Ⰷ N and 2V2N Ⰷ DRp 共t兲 / t agrees qualitatively with results reported by Lobaskin et al. 关33兴 for triangular swimmers in the moderate Reynolds number regime R ⬃ 1. In this context, we also mention recent work by Golestanian et al. 关50兴, who discuss similar scaling relations for the diffusion of phoretic swimmers. In the remainder, we are going to compare the analytical predictions with results of computer simulations, based on a direct numerical integration of the Langevin Eq. 共5兲 for the spheres. Our main focus is on the transition from the passive Brownian motions to the active swimming regime. 冉 冊 冉 冊 冉 冊 7 2 ᐉ a 2 sin ⌬ = 0.292 , 24 ᐉ s 16ᐉ3 8ᐉ −1 k BT 3a ⯝ −1 = 0.473 ᐉ 3s , m3 1 k BT 3 5 m3 = 5.127 + , 9 a ᐉ ᐉs 共23a兲 共23b兲 共23c兲 and 2V2N = 0.080 ᐉ5 . m 3s 共23d兲 It is remarkable that increasing the swimmer size by one order of magnitude increases the orientation correlation time N by three orders of magnitude. Figures 1 and 2 depict the results of numerical simulations 共symbols兲 of the Langevin Eq. 共5兲 and also the corresponding theoretical predictions 共dashed lines兲. The numerical data points represent averages over 100 trajectories with identical initial conditions. More precisely, at time t = 0 the swimmer is pointing along the x3 axis with the first sphere being located at the origin, i.e., X1共0兲 = 0, X2共0兲 = ᐉN共0兲, and X3共0兲 = 关2ᐉ − sin共⌬兲兴N共0兲, where N共0兲 = 共0 , 0 , 1兲. As evident from the diagrams in Figs. 1 and 2, the results of the numerical simulations are very well matched by the theoretical curves over several orders of magnitude in time. In particular, for a leg length in the range ᐉ ⬃ 5 m 共red “+”/green “⫻” symbols兲 one readily observes the three aforementioned regimes: 共i兲 Brownian diffusion at small time scales t Ⰶ N, 共ii兲 ballistic behavior at intermediate time scales, and 共iii兲 quasidiffusive behavior due to noise-induced rotation for t Ⰷ N. We conclude the discussion of the numerical results by addressing a few technical aspects that might be relevant and helpful for future simulations. When considering ensembles 021903-5 PHYSICAL REVIEW E 80, 021903 共2009兲 DN (t) = N (t)N (0) JÖRN DUNKEL AND IRWIN M. ZAID displacement and other statistical observables agree with those obtained for very small time steps dt = 0.1T␥. Generally, a satisfactory resolution of the bending and relaxation dynamics of the legs/spheres would require dt Ⰶ 共M / K兲1/2 and dt Ⰶ 共M / k0兲1/2, respectively. 1 0.5 ◦ = 1.0 µm + = 2.5 µm × = 5.0 µm = 10.0 µm 0 0.01 0.1 1 10 100 V. CONCLUSIONS 1000 t [s] FIG. 1. 共Color online兲 Orientation correlation function for four different swimmer sizes 2ᐉ. Symbols represent averages over 100 trajectories, numerically calculated from the Langevin Eq. 共5兲 using parameters as described in the text. The dashed lines depict the theoretically predicted exponential decay DN共t兲 ⯝ exp共−t / N兲 with orientation correlation time N determined by Eq. 共15兲. It is remarkable that changing the swimmer size by one order of magnitude increases the correlation time by three orders of magnitude. with N ⬎ 1 swimmers the computationally most expensive step is the Cholesky decomposition of the diffusion tensor, see Eq. 共3兲, which is approximately of the order O关共pN兲3兴 关36兴. It is also worthwhile to briefly comment on the choice of the time step dt in the Langevin simulations. Ideally, one would like to choose dt smaller than the smallest dynamical time scale in the system, which for our model is given by the damping time T␥ = M / 共6a兲. For the swimmer parameters considered here we find T␥ ⬃ 2 ⫻ 10−9共ᐉ / m兲2s, which means that adopting dt ⬃ T␥ would not allow us to simulate experimentally accessible time scales in the seconds range. Since we are not interested in the dynamical details at very short times scales, we choose in our simulations the time step larger than T␥, but much smaller than the period T of a swimming stroke by fixing dt = 10−3T for ᐉ ⬍ 5 m and dt = 10−2T if ᐉ ⱖ 5 m. We verified, however, that for intermediate time scales 共of the order of a few stroke periods T ⬃ 6 ⫻ 10−3 s兲 the numerical results for the mean square DR (t)/t [µm2 /s] 104 103 102 ◦ = 1.0 µm + = 2.5 µm × = 5.0 µm = 10.0 µm 101 Understanding the interplay between Brownian motion, hydrodynamic interactions, and self-propulsion is a prerequisite for understanding the dynamics of bacteria and artificial swimming devices at the microscale. In the first part, we discussed how one can model these phenomena by means of stochastic processes 共overdamped Langevin equations兲. Subsequently, as a first application, we focused on the size dependence of diffusive behavior at low Reynolds numbers 共R Ⰶ 1兲 for a quasilinear three-sphere swimmer model. Our theoretical analysis complements a recent experimental study by Howse et al. 关6兴, who investigated the diffusion of chemically driven, spherical colloids, and theoretical work by Lobaskin et al. 关33兴, who considered the Brownian dynamics of an artificial triangular microswimmer at moderate Reynolds numbers 共R ⬃ 1兲. Starting from the Kirkwood-Smoluchowski equation 关34–37兴, we derived analytical results for the orientation correlation time, the mean velocity, and the mean-square displacement of an overdamped, quasilinear three-sphere swimmer 关20,25兴. Analytical formulae as derived here are useful for testing numerical simulations and 共in兲 validating simplified probabilistic models 关39兴. Moreover, they provide detailed insight into the size-regulated transition from predominantly random to quasiballistic motions. The proposed method of modeling swimmers by effective potentials within a Langevin scheme can be readily extended to study complex behavior in larger swimmer ensembles. However, at high swimmer densities, collisions, and nearfield hydrodynamics affect the diffusive behavior 关47,51,52兴 and it will be necessary to modify the hydrodynamic interaction tensor accordingly. Generally, a useful dimensionless quantifier for the efficiency of active swimming relative to diffusion is given by the ratio ⑀= 100 10 −1 0.001 0.01 0.1 1 10 100 1000 t [s] FIG. 2. 共Color online兲 Mean-square displacement DR共t兲 ª 具关R共t兲 − R共0兲兴2典 divided by time t for the same set of parameters as in Fig. 1. The dashed lines correspond to the analytical formula 共21兲. The dynamics of small swimmers 共ᐉ = 1 m, blue circles兲 is dominated by Brownian motion on all time scales, whereas big swimmers 共ᐉ = 10 m, black diamonds兲 can move ballistically for several minutes. In the intermediate region 共ᐉ ⬃ 5 m, red “+“/green “⫻“兲 we observe a ballistic transition from ordinary Brownian motion at small times t Ⰶ N to noise-induced rotational diffusion at large times t Ⰷ N. Since N ⬃ ᐉ3, the transition from Brownian to quasiballistic motion is very sharp. Interestingly enough, typical sizes of bacteria lie in or near this niche 关10兴. N . ⌳/V 共24兲 The denominator corresponds to the time needed by a unperturbed swimmer of velocity V to move one body length ⌳, and the swimmer geometry is encoded in the orientation correlation time N. For ⑀ Ⰷ 1 共⑀ Ⰶ 1兲 self-propulsion is effective 共noneffective兲. For nonisolated swimmers, N is not only determined by thermal effects, but also by collisions with other swimmers 关51,52兴. With regard to future studies we note that the combination of thermal fluctuations and hydrodynamic coupling might also lead to interesting behavior in simple arrangements of microswimmers. For example, experiments have shown that colloidal spheres localized in an array of optical traps create memory effects 关53兴 and driven vibrations 关54兴. Our formalism provides a starting point for the investigation of many 021903-6 PHYSICAL REVIEW E 80, 021903 共2009兲 NOISY SWIMMING AT LOW REYNOLDS NUMBERS self-propelled bodies in separate potentials, which could be helpful for interpreting experimental data of trapped bacteria or for constructing pumps from a collection of microswimmers. To summarize, the above results may provide guidance for constructing artificial microswimmers or pumps that work efficiently in the critical transition region that separates Brownian from quasideterministic motions. In particular, by tuning the parameters to the narrow cross-over region one could construct swimmers that explore with high probability a maximized volume fraction within a given period of time. The fact that many bacteria live near or exactly in this niche 关10兴 suggest that this may indeed be a useful strategy. Thus, exploiting the interplay between noise and active self-motion could lead to novel applications 关55兴, e.g., with regard to the controlled transport 关56兴 and distribution of chemical and biological substances in small scale technical devices or even within the human body. H共O␣i兲共 j兲 ª ⯝ The authors would like to thank Matthieu Dufay, Bortolo Mognetti, Olivier Pierre-Louis, Victor Putz, and Julia Yeomans for helpful discussions. This work was supported by the EPSRC grant no. EP/D050952/1 共J.D.兲. APPENDIX: CALCULATIONS This appendix provides derivations of analytical results for the orientation correlation function, the mean velocity, and the spatial mean square displacement from the Langevin Eq. 共5兲. Our calculations are based on the following simplifying assumptions: 共i兲 The motion of the three-sphere swimmer is approximately stiff and shape-driven, i.e., 共A1a兲 d23 ª X3 − X2 ⯝ Nd23 , 共A1b兲 d13 ª X3 − X1 ⯝ N共d12 + d23兲, 共A1c兲 d12 = ᐉ + 12 sin共t兲, 共A1d兲 d23 = ᐉ + 23 sin共t − ⌬兲. 共A1e兲 共A2a兲 and, with F3i + F2i + F1i = 0, F2i = − 共g1 + g3兲Ni , 共A2b兲 where g␣ is the force amplitude, and Ni denotes a component of the orientation vector N共t兲 ª X3 − X1 . 兩X3 − X1兩 1 共␦ij + NiN j兲. 8d␣ 共A3兲 1. Orientation correlation function Given a deterministic initial state N共0兲 = 关Nk共0兲兴, we are interested in DN共t兲 ª 具N共t兲N共0兲典 = 具Nk共t兲典Nk共0兲. The first step is to find the stochastic differential equation 共SDE兲 for the orientation vector N共t兲. This can be achieved by virtue of the Ito formula 关49,57兴, yielding 共A2c兲 共A4兲 where dX共␣i兲共t兲 is governed by Eq. 共5兲, and the diffusion tensor is given by D ª kBTH ⯝ kBTHO. To determine 具Nk共t兲典, we use that 具C共␣i兲共␥k兲dB共␥k兲共t兲典 = 0 for an Ito SDE, and therefore d具X共␣i兲共t兲典 = 具H共␣i兲共 j兲F共 j兲典dt. 共A5兲 Taking the average of Eq. 共A4兲 and inserting Eq. 共A5兲, we obtain 具Ṅk共t兲典 = 具H共␣i兲共 j兲F共 j兲共␣i兲Nk典 + 具D共␣i兲共 j兲共␣i兲共 j兲Nk典, 共A6兲 where d具Ṅk共t兲典 ª 具dNk共t兲 / dt典. We next evaluate the two terms the right-hand side of Eq. 共A6兲 separately. To this end we note that 共␣ j兲Nk = 1 共␦3␣ − ␦1␣兲共␦ jk − N jNk兲, d13 共A7兲 yielding for the first term H共␣i兲共 j兲F共 j兲共␣i兲Nk = 关H共3i兲共 j兲 − H共1i兲共 j兲兴 ⫻ F共 j兲 d13 共␦ik − NiNk兲. Using Oseen approximation Eq. 共A3兲, we find In this case the internal forces, which generate the swimming strokes, point along the swimmer’s axis and we may write F3i = g3Ni 冊 The second line follows from assumption Eq. 共A1兲. where F1i = g1Ni, 冉 1 d d ␦ij + ␣i2 ␣ j , 8d␣ d␣ dNk共t兲 = 共␣i兲NkdX共␣i兲 + D共␣i兲共 j兲共␣i兲共 j兲Nkdt, ACKNOWLEDGMENTS d12 ª X2 − X1 ⯝ Nd12 , 共ii兲 We adopt the Oseen approximation, i.e., H ⯝ HO where H共3i兲共 j兲F共 j兲 = 冋 共A8兲 册 共g3 + g1兲 g3 g1 + − N, ␥3 4d31 4d32 i and, similarly, H共1i兲共 j兲F共 j兲 ⬀ Ni. Taking into account that Ni共␦ik − NiNk兲 ⬅ 0, 共A9兲 the first term on the right-hand side of Eq. 共A6兲 vanishes, and Eq. 共A6兲 reduces to 具Ṅk共t兲典 = 具D共␣i兲共 j兲共␣i兲共 j兲Nk典. By virtue of Eq. 共A7兲, we find 021903-7 共A10兲 PHYSICAL REVIEW E 80, 021903 共2009兲 JÖRN DUNKEL AND IRWIN M. ZAID 1 共␣i兲共 j兲Nk = 2 d13 3 1 具Ṙi共t兲典 = 兺 具H共␣i兲共 j兲F共 j兲典 3 ␣=1 共␦3 − ␦1兲共␦3␣ − ␦1␣兲 ⫻ 共3N jNkNi − N j␦ik − Nk␦ij − Ni␦ jk兲. 3 共A11兲 To obtain the right-hand side of Eq. 共A10兲, we still need to contract with the diffusion tensor D共␣i兲共 j兲. This results in two contributions: From the diagonal part we get 1 共␦3␣ − ␦1␣兲共␦3␣ − ␦1␣兲 ␥ 共␣ j兲 ␣ k BT 兺 ⫻ 1 2 d13 = − k BT 冊 1 2Nk 1 + 2 , ␥1 ␥3 d13 共A12a兲 3 2d13 冓冉 1 g1 g2 g3 g2 g3 + + + + Ni 3 ␥1 ␥2 ␥3 4d12 4d13 冊 g1 g3 g1 g2 + + + 4d12 4d23 4d13 4d23 冋 ⫻ 册 k BT 1 2 1 + 具Nk典, 2 + 3 ␥1 ␥3 d13 2d13 ¬ 2. Motion of the geometric center 冋 冉 1 1 1 1 1 − + − ␥3 ␥2 4 d13 d12 冊册 冊册 1 + 具Nig3典 3 A1 A3 具Nig1典 + 具Nig3典. 3 3 Following a similar procedure, we can derive analytical expression for the mean velocity 具Ṙ共t兲典 and the mean-square displacement DR共t兲. a. Mean velocity First, we would like to determine the mean velocity 具Ṙ共t兲典 of the swimmer’s geometric center 共A17兲 具ḋ12k典 = 具Ṅk典d12 + 具Nk典ḋ12 , 共A18a兲 具ḋ23k典 = 具Ṅk典d23 + 具Nk典ḋ23 . 共A18b兲 On the other hand, from the definition of the vectors d␣ and the Langevin equations for dX␣, we have 共A13兲 具ḋ12k典 = 具H共2k兲共 j兲F共 j兲 − H共1k兲共 j兲F共 j兲典, 共A19a兲 具ḋ23k典 = 具H共3k兲共 j兲F共 j兲 − H共2k兲共 j兲F共 j兲典. 共A19b兲 Inserting the explicit expressions for H共␣i兲共 j兲 and F共 j兲, Eqs. 共A19兲 can be rewritten as Averaging the stochastic differential equation 1 dR共t兲 ª 共dX1 + dX2 + dX3兲 3 冊冔 Consequently, to find 具Ṙi共t兲典, we still need to determine the mean forces 具F共i兲典 = 具Nig典 on the first and last sphere,  = 1 , 3. This can be achieved as follows: Eq. 共A1兲 implies that which for ␥␣ = 6a␣ gives Eq. 共14兲. 1 R共t兲 ª 共X1 + X2 + X3兲. 3 冉 1 1 1 1 1 1 具Ṙi共t兲典 = 具Nig1典 − + − 3 ␥1 ␥2 4 d13 d23 Combining the two contributions we find 冋 冉 共␦ij + NiN j兲 g g N j = N i 兺 . 8d␣ ⫽␣ 4d␣ Since the internal swimming forces sum to zero, we may eliminate g2 by using g2 = −共g1 + g3兲 to obtain 共A12b兲 具Ṅk共t兲典 = − kBT 冔 + 共1 − ␦␣兲H共␣i兲共 j兲F共 j兲 . Inserting this into Eq. 共A15兲 gives + Nk . ␥␣ Considering as before a stiff, shape-driven swimmer and Oseen interactions H = HO, we have F共i兲 ⯝ gNi and, therefore, 具Ṙi共t兲典 = 共␦ij + NiN j兲 1 ⫻ 2 共3NiNkN j − N j␦ik − Nk␦ij − Ni␦ jk兲 8d␣ d13 = F共␣i兲 共A16兲 兺 共1 − ␦␣兲共␦3␣ − ␦1␣兲共␦3 − ␦1兲 共␣i兲共 j兲 k BT 冓 共A15兲 while the hydrodynamic off-diagonal terms give k BT 1 兺 3 ␣=1 兺 H共␣i兲共 j兲F共 j兲 ⯝ 兺 ⫽␣ ⫽␣ 共3N jNkN j − N j␦ jk − Nk␦ jj − N j␦ jk兲 冉 = 共A14兲 具ḋ12k典 = − with respect to the underlying Wiener process and dividing by dt, we obtain − 021903-8 冋 冋 册 1 1 1 + − 具Nkg1典 ␥1 ␥2 2d12 冉 1 1 1 1 1 − + − ␥2 4 d12 d23 d13 冊册 具Nkg3典 PHYSICAL REVIEW E 80, 021903 共2009兲 NOISY SWIMMING AT LOW REYNOLDS NUMBERS ¬− B1具Nkg1典 − C具Nkg3典 and 具ḋ23k典 = 冋 冉 共A20a兲 1 1 1 1 1 − + − ␥2 4 d12 d23 d13 冋 册 冊册 1 1 1 + − 具Nkg3典 + ␥2 ␥3 2d23 共A20b兲 具ḋ12k典 = − B1具Nkg1典 − C具Nkg3典, 共A21a兲 具ḋ23k典 = C具Nkg1典 + B3具Nkg3典, 共A21b兲 With left-hand side given by Eqs. 共A19兲. This is easily done and we may summarize the result for the mean velocity A1 A3 具Nkg1典 + 具Nkg3典, 3 3 共A22a兲 where B3具ḋ12k典 + C具ḋ23k典 , 具Nkg1典 = − B 1B 3 − C 2 dsV共s兲, 共A25兲 具Ṙk共t兲典 ⯝ V具Nk共t兲典. 共A26兲 t For example, when considering equal-sized beads with a␣ = a and ᐉ Ⰷ max兵a , 12 , 23其, then Hence, in order to obtain the unknown expectation values 具Nkg1典, we have to solve the linear system 具Ṙk共t兲典 = V共t兲 = a23 a12 cos共t + 12兲 + cos共t + 23兲 + V, ᐉ ᐉ where 冉 冊 7 1223 a sin ⌬ 24 ᐉ2 V= C具ḋ12k典 + B1具ḋ23k典 , B 1B 3 − C 2 b. Spatial diffusion Using the result for dR共t兲 from above, we may rewrite the mean square displacement DR共t兲 as DR共t兲 ª 具关R共t兲 − R共0兲兴2典 共A22b兲 = 冓冕 t dRk共u兲 = 冕 冕 冓冋 t ds 0 冋 ⫻ 共A22d兲 du 0 共A22e兲 Since the quantities 具Nk典, 具Ṅk典, d␣, ḋ␣, A␣, B␣, and C are known we have thus obtained a closed analytical result for the mean swimmer velocity within the Oseen approximation. Analogous calculations can be performed for HM , but do not yield much additional insight 共for a single swimmer兲. Additional simplifications. If the orientation correlation time N is larger than the driving period T = 2 / then 具ḋ12k典 ⯝ 具Nk典ḋ12, 具ḋ23k典 ⯝ 具Nk典ḋ23 . ⬘ 冉 A1 A3 N kg 1 + N kg 3 3 3 冕 册冔 册 s u 具关C共␣k兲共␥n兲兴sdB共␥n兲共s兲 共A28兲 Here, we have again used that 具兰f共X␣兲dB共␥n兲共t兲典 = 0 holds for Ito integrals. We consider the two remaining integrals in Eq. 共A28兲 separately, starting with the second one. We find 3 DRp 共t兲 ª 共A23兲 1 兺 9 ␣,␣ =1 ⬘ 冕 具关C共␣k兲共␥n兲兴sdB共␥n兲共s兲 ⫻ 关C共␣⬘k兲共␥⬘n⬘兲兴udB共␥⬘n⬘兲共u兲典 具Ṙk共t兲典 ⯝ V共t兲具Nk共t兲典, V共t兲 = − 冔 ⫻ 关C共␣⬘k兲共␥⬘n⬘兲兴udB共␥⬘n⬘兲共u兲典. In this case, we may simplify where 1 兺 9 ␣,␣ =1 + dRk共s兲 A1 A3 N kg 1 + N kg 3 3 3 3 具ḋ23k典 = 具Ṅk典d23 + 具Nk典ḋ23 . 冕 t 0 t 共A22c兲 with 具ḋ12k典 = 具Ṅk典d12 + 具Nk典ḋ12 , 共A27兲 and ⌬ ª 12 − 23, and higher-order terms have been neglected. 0 具Nkg3典 = t+T so that 具Nkg1典 ¬C具Nkg1典 + B3具Nkg3典. 冕 Vª 冊 冉 A3 Cḋ12 + B1ḋ23 A1 B3ḋ12 + Cḋ23 + 3 3 B 1B 3 − C 2 B 1B 3 − C 2 3 共A24a兲 = 2 兺 9 ␣,␣ =1 ⬘ 冊 2 = k BT 9 共A24b兲 is a periodic function, V共t兲 = V共t + T兲. Since we assumed N Ⰷ T, we can achieve further simplification by replacing V共t兲 with its stroke average 冕 t ds具D共␣k兲共␣⬘k兲典 0 冕 冋 t ds 0 3 3 3 3 + + ␥1 ␥2 ␥3 2 + kBT 兺 共1 − ␦␣␣⬘兲 9 ␣,␣ =1 ⬘ where 021903-9 册 冕 t 0 O dsH共␣k兲共␣⬘k兲 , PHYSICAL REVIEW E 80, 021903 共2009兲 JÖRN DUNKEL AND IRWIN M. ZAID O H共␣k兲共␣⬘k兲 = 1 ␦kk + NkNk = . 8d␣␣⬘ 2d␣␣⬘ 共A29兲 For spherical particles we have ␥␣ = 6a␣ and, therefore, 冉 冊 1 k BT 1 1 1 + + t DRp 共t兲 = 9 a1 a2 a3 2 k BT 9 + 冕 冉 t ds 0 relation functions. However, instead of calculating these correlation functions exactly, we may approximate, for N Ⰷ T, the integrand by 关cf. Eqs. 共A17兲 and 共A26兲兴 具关 . . . 兴s关 . . . 兴u典 ⯝ V2具Nk共s兲Nk共u兲典, where V is the stroke-averaged velocity of the corresponding deterministic swimmer, cf. Eqs. 共A25兲 and 共A26兲. Adopting this approximation we find 冊 1 1 1 + + . d12 d23 d13 t For ᐉ Ⰷ max兵␣其, the integrand in the second line can be approximated by 5 / 共2ᐉ兲 yielding DRp 共t兲 冉 冊 1 k BT 1 1 1 5 ⯝ + + + t. 9 a1 a2 a3 ᐉ 冕 冕 冓冋 t DRa 共t兲 ª ds 0 冋 du 0 A1 A3 N kg 1 + N kg 3 3 3 A1 A3 N kg 1 + N kg 3 ⫻ 3 3 册冔 册 t ds 0 du具Nk共s兲Nk共u兲典 0 t = V2 t ds 0 共A30兲 It remains to determine the first 共double兲 integral in Eq. 共A28兲, reading t 冕 冕 冕 冕 DRa 共t兲 ⯝ V2 du exp共− 兩u − s兩/N兲 0 = 2V2N关t + N共e−t/N − 1兲兴, and thus the final result DR共t兲 = DRp 共t兲 + DRa 共t兲 s ⯝ . u 冉 冊 1 1 5 1 k BT 1 + + + t 9 a1 a2 a3 ᐉ + 2V2N关t + N共e−t/N − 1兲兴. 共A31兲 The subscripts indicate the time arguments in the bracketed expressions, respectively. Upon recalling that F共␣i兲 = g␣Ni is the internal force acting on sphere ␣, we see that the contribution DRa 共t兲 is essentially determined by the force-force cor- The first part represents passive 共thermal兲 diffusion; the second part is due to active swimming 共note that N is temperature dependent as well兲. 关1兴 R. Dreyfus, J. Baudry, M. L. Roper, M. Fermigier, H. A. Stone, and J. Bibette, Nature 共London兲 437, 862 共2005兲. 关2兴 A. Sokolov, I. S. Aranson, J. O. Kessler, and R. E. Goldstein, Phys. Rev. Lett. 98, 158102 共2007兲. 关3兴 A. P. Berke, L. Turner, H. C. Berg, and E. Lauga, Phys. Rev. Lett. 101, 038102 共2008兲. 关4兴 M. Leoni, J. Kotar, B. Bassetti, P. Cicuta, and M. C. Lagomarsino, Soft Matter 5, 472 共2009兲. 关5兴 P. Tierno, R. Golestanian, I. Pagonabarraga, and F. Sagues, J. Phys. Chem. B. 112, 16525 共2008兲. 关6兴 J. R. Howse, R. A. L. Jones, A. J. Ryan, T. Gough, R. Vafabakhsh, and R. Golestanian, Phys. Rev. Lett. 99, 048102 共2007兲. 关7兴 P. Tierno, R. Golestanian, I. Pagonabarraga, and F. Sagues, Phys. Rev. Lett. 101, 218304 共2008兲. 关8兴 A. Shapere and F. Wilczek, J. Fluid Mech. 198, 557 共1989兲. 关9兴 H. A. Stone and A. D. T. Samuel, Phys. Rev. Lett. 77, 4102 共1996兲. 关10兴 M. F. Copeland and D. B. Weibel, Soft Matter 5, 1174 共2009兲. 关11兴 O. Reynolds, Philos. Trans. R. Soc. Lond. 174, 935 共1883兲. 关12兴 J. Lighthill, SIAM Rev. 18, 161 共1976兲. 关13兴 E. M. Purcell, Am. J. Phys. 45, 3 共1977兲. 关14兴 J. Koiller, K. Ehlers, and R. Montgomery, J. Nonlinear Sci. 6, 507 共1996兲. 关15兴 E. Yariv, J. Fluid Mech. 550, 139 共2006兲. 关16兴 E. Lauga and D. Bartolo, Phys. Rev. E 78, 030901共R兲 共2008兲. 关17兴 H. C. Berg, Phys. Today 53共1兲, 24 共2000兲. 关18兴 C. Dombrowski, L. Cisneros, S. Chatkaew, R. E. Goldstein, and J. O. Kessler, Phys. Rev. Lett. 93, 098103 共2004兲. 关19兴 L. E. Becker, S. A. Koehler, and H. A. Stone, J. Fluid Mech. 490, 15 共2003兲. 关20兴 A. Najafi and R. Golestanian, Phys. Rev. E 69, 062901 共2004兲. 关21兴 T. Ishikawa, G. Sekiya, Y. Imai, and T. Yamaguchi, Biophys. J. 93, 2217 共2007兲. 关22兴 D. J. Earl, C. M. Pooley, J. F. Ryder, I. Bredberg, and J. M. Yeomans, J. Chem. Phys. 126, 064703 共2007兲. 关23兴 C. M. Pooley, G. P. Alexander, and J. M. Yeomans, Phys. Rev. Lett. 99, 228103 共2007兲. 关24兴 G. P. Alexander and J. Yeomans, Europhys. Lett. 83, 34006 共2008兲. 关25兴 R. Golestanian and A. Ajdari, Phys. Rev. E 77, 036308 共2008兲. 关26兴 Y. Yang, J. Elgeti, and G. Gompper, Phys. Rev. E 78, 061903 共2008兲. 关27兴 E. E. Keaveny and M. R. Maxey, Phys. Rev. E 77, 041910 共2008兲. 关28兴 P. T. Underhill, J. P. Hernandez-Ortiz, and M. D. Graham, Phys. Rev. Lett. 100, 248101 共2008兲. 关29兴 U. Erdmann and W. Ebeling, Fluct. Noise Lett. 3, L145 共2003兲. 关30兴 A. Baskaran and M. C. Marchetti, Phys. Rev. Lett. 101, 268101 共2008兲. 关31兴 A. Baule, K. V. Kumar, and S. Ramaswamy, J. Stat. Mech.: Theory Exp. 共2008兲 P11008. 关32兴 S. van Teeffelen and H. Löwen, Phys. Rev. E 78, 020101共R兲 共2008兲. 021903-10 PHYSICAL REVIEW E 80, 021903 共2009兲 NOISY SWIMMING AT LOW REYNOLDS NUMBERS 关33兴 V. Lobaskin, D. Lobaskin, and I. Kulic, Eur. J. Phys. Spec. Top. 157, 149 共2008兲. 关34兴 T. J. Murphy and J. L. Aguirre, J. Chem. Phys. 57, 2098 共1972兲. 关35兴 M. Bixon, Annu. Rev. Phys. Chem. 27, 65 共1976兲. 关36兴 B. Dünweg and J. C. Ladd, Advanced Computer Simulation Approaches for Soft Matter Sciences III 共Springer, Berlin, Heidelberg, 2009兲, Vol. 221, pp. 89–166. 关37兴 B. Liu and B. Dünweg, J. Chem. Phys. 118, 8061 共2003兲. 关38兴 B. Guirao and J. F. Joanny, Biophys. J. 92, 1900 共2007兲. 关39兴 P. S. Lovely and F. W. Dahlquist, J. Theor. Biol. 50, 477 共1975兲. 关40兴 C. W. Oseen, Neuere Methoden und Ergebnisse in der Hydrodynamik 共Akademischer Verlag, Leipzig, 1927兲. 关41兴 J. Happel and H. Brenner, Low Reynolds Number Hydrodynamics, International Series in the Physical and Chemical Engineering Sciences 共Prentice-Hall, Inc., Englewood Cliffs, N.J., 1965兲. 关42兴 R. Zwanzig, J. Kiefer, and G. H. Weiss, Proc. Natl. Acad. Sci. U.S.A. 60, 381 共1968兲. 关43兴 J. Rotne and S. Prager, J. Chem. Phys. 50, 4831 共1969兲. 关44兴 关45兴 关46兴 关47兴 关48兴 关49兴 关50兴 关51兴 关52兴 关53兴 关54兴 关55兴 关56兴 关57兴 021903-11 P. Mazur, Physica 110A, 128 共1982兲. P. Mazur and W. van Saarloos, Physica 115A, 21 共1982兲. H. Yamakawa, J. Chem. Phys. 53, 436 共1970兲. T. Ishikawa, M. P. Simmonds, and T. J. Pedley, J. Fluid Mech. 568, 119 共2006兲. E. Wajnryb, P. Szymczak, and B. Cichocki, Physica A 335, 339 共2004兲. P. Hänggi and H. Thomas, Phys. Rep. 88, 207 共1982兲. R. Golestanian, T. B. Liverpool, and A. Ajdari, New J. Phys. 9, 126 共2007兲. T. Ishikawa and T. J. Pedley, J. Fluid Mech. 588, 437 共2007兲. T. Ishikawa, J. T. Locsei, and T. J. Pedley, J. Fluid Mech. 615, 401 共2008兲. J.-C. Meiners and S. R. Quake, Phys. Rev. Lett. 82, 2211 共1999兲. M. Polin, D. G. Grier, and S. R. Quake, Phys. Rev. Lett. 96, 088101 共2006兲. R. D. Astumian and P. Hänggi, Phys. Today 55共11兲, 33 共2002兲. R. Golestanian, Eur. Phys. J. E 25, 1 共2008兲. J. Dunkel and P. Hänggi, Phys. Rep. 471, 1 共2009兲.