Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Flow structures govern particle collisions in turbulence

Flow structures govern particle collisions in turbulence 1, 2,y 1,z 1,x Jason R. Picardo, Lokahith Agasthya, Rama Govindarajan, and Samriddhi Sankar Ray International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bangalore 560089, India Indian Institute for Science Education and Research, Pune 411008, India The role of the spatial structure of a turbulent ow in enhancing particle collision rates in suspensions is an open question. We show and quantify, as a function of particle inertia, the correlation between the multiscale structures of turbulence and particle collisions: Straining zones contribute predominantly to rapid head-on collisions compared to vortical regions. We also discover the importance of vortex-strain worm-rolls, which goes beyond ideas of pref- erential concentration and may explain the rapid growth of aggregates in natural processes, such as the initiation of rain in warm clouds. Turbulence is riddled with a hierarchy of interacting vortical and straining structures (Fig. 1), which are closely related to its characteristic intermittent and non-Gaussian statistics [1{6]. The most intense structures typically occur near each other, in the form of vortex tubes surrounded by straining sheets, as shown in Fig. 1. This organization|a sort of vortex-strain worm-rolls |is characteristic of turbulent ows [7{10], and its origin and dynamical implications continue to be investigated [11{16]. These structures distinguish fully developed turbulence from purely random ow elds, and must play an important role in many aspects of turbulent transport. The most important of these|because it remains central to our understanding of phenomena as diverse as the formation of planets in circumstellar disks [17] or the initiation of rain in warm clouds [18, 19]| is the growth of macroscopic aggregates, due to collisions and coalescences, from nuclei-particles (dust or aerosols) suspended in a turbulent ow. The role of the underlying turbulent carrier ow is critical: Estimates of, e.g., the rate of growth of these aggregates in the absence of such ows do not agree with that seen in nature [20]. Indeed, the explanation of such rapid growth through coalescence, demonstrated [21{23] and quanti ed in terms of ow statistics [20, 24, 25], is rooted in the ability of turbulent ows to enhance the rate of collisions between nuclei seed-particles. A critical discovery, due to Bec, et al. [22], was to nd the precise connection between the inter- mittent (multiscaling) nature of the carrier turbulent ow and the accelerated growth of aggregates. And yet the implied correlation between the structure of the ow and droplet collisions-coalescences remains unknown. Indeed, there is evidence to suggest that ow structures matter. But the ques- tion is how and when? In this paper we answer this question comprehensively, based on direct numerical simulations (DNSs), and show how straining regions are intrinsically more e ective at generating collisions than vortical ones even for uniformly distributed inertia-less particles. Particle inertia widens this discrepancy, not simply by preferential concentration, but also by selectively increasing the collision velocities in straining zones. This is because straining regions have a larger proportion of head-on or rear-end collisions, as opposed to side-on collisions, which are predominant in vortical regions. Consequently, a larger fraction of the velocity gradient in straining zones is translated into the particle approach velocity. Finally, and most strikingly, we show how intense vorticity and strain, cohabiting as vortex-strain worm-rolls, conspire to generate rapid, violent collisions. We therefore consider an incompressible (ru = 0) turbulent ow whose velocity u is a solution to the Navier-Stokes equation @ u + (ur)u = rp + r u + f (1) jrpicardo@icts.res.in; picardo21@gmail.com lokahith.agasthya@students.iiserpune.ac.in rama@icts.res.in samriddhisankarray@gmail.com arXiv:1810.10285v2 [physics.flu-dyn] 26 Mar 2019 2 FIG. 1. A representative snapshot of three-dimensional contours of Q showing intense vortex tubes (opaque p p 2 2 red: +5:6 hQ i) enveloped by dissipative, straining sheets (transparent blue: 2 hQ i). 3 3 where  is the kinematic viscosity. We perform DNSs on a tri-periodic domain with N = 512 grid points, by using a standard de-aliased pseudo-spectral solver [26] and a second-order Adams- Bashford scheme for time-integration. A statistically stationary, homogeneous and isotropic ow is maintained by the time-dependent large-scale forcing f , which injects a constant amount of energy, and hence dissipation , into the rst two wavenumber shells. The Kolmogorov length 3 1=4 = ( =) satis es k  1:7 (where k = 2N=3 is the maximum resolved wavenumber). max max The Taylor-Reynolds number Re = 2E 5=(3) = 196, where E is the total kinetic energy. The compensated energy spectrum from such a simulation shows an inertial range which is certainly less than a decade but still resolved (see, e.g., Fig. 1 (a) for N = 512 and D = 3:00 in Ref. [5]). To identify vortical and straining structures in the ow, we use the second invariant of the 2 2 velocity gradient tensor Q = (R S )=2 [27{29], de ned through the velocity gradient tensor 1=2 A =  ru (normalized by the Kolmogorov time  = (=) ) which yields the symmetric strain T T rate tensor S = (A +A )=2 and the anti-symmetric rotation rate tensor R = (AA )=2. Regions where Q > 0 (Q < 0) are dominated by vorticity (irrotational strain) [25]. Figure 1 presents contours of large positive and negative values of Q, which reveal characteristic vortex-strain worm- rolls. We introduce in the ow 10 identical particles, each having a sub-Kolmogorov radius a = =3 and a density  much larger than that of the carrier- uid  . The particles occupy a small volume p f fraction of O(10 ) and their in uence on the ow is negligible. Since the Reynolds number associated with their slip velocities is small, and    , the evolution of particle trajectories p f X (t) is determined by the simpli ed Maxey-Riley equations [30{33]: dX dV 1 p p = V ; = [V u(X ; t)] (2) p p p dt dt p 3 where  = 2a  =(9 ), the particle relaxation time, yields the Stokes number (St =  = ), p p f p which provides a non-dimensional measure of the particle's inertia. We consider several families of particles, with St ranging from 0 to 16.75 and use an exponential integration scheme [34]. The case of tracers St = 0 is handled separately by using a second order Runge-Kutta time-stepper. The uid velocity at the particle position is obtained via fourth-order B-spline interpolation [35]. It is important to note that the particle radius in uences particle transport in two distinct ways: It sets the value of St, while also determining the separation between particle centers at collision. As we are considering heavy particles ( = > 1), the radius must be very small for small values of St. This makes the gathering of collision statistics in this dilute suspension increasingly impractical as St ! 0, unless we consider an arti cially enlarged collision radius. Indeed, for tracers, by de nition, any choice of radius is arbitrary. Our choice of =3 is physically consistent for particles with St > 0:1. However, for smaller St, this particle radius would imply  = < 1 p f which contradicts our heavy particle assumption. Indeed, to accurately describe particle motion, the simpli ed Maxey-Riley equations (2) require  = > O(10 ) [31{33]. Nevertheless, it is p f convenient to consider this enlarged collision radius, solely for the purpose of detecting collisions. As a check, we actually performed simulations with particle radius =10 (where the problem is mitigated) and found our results unchanged. This can be rationalised by noting that particles move ballistically in the small distance which separates the =3 and the more realistic =10 radii and hence make our collision results insensitive to the precise choice of the radius. Nevertheless, we do report results for the larger radius because the collision statistics are much poorer with the smaller radius. After the randomly-seeded particles have settled into a statistically stationary distribution, we begin detecting collisions using an algorithm similar to that in Ref. [21]. After a collision is detected, the two particles are allowed to move past each other without any modi cation to their trajectories. This ghost collision approach is a standard approximation that, by ignoring coalescence, allows one to measure collision rates while the particle distribution remains in a statistically stationary state. This feature allows us to identify and compare the regions of the ow where most of the collisions occur to those regions where most of the particles reside. Of course, because particles never coalesce, the collision rates are over-predicted (this issue is less severe for dilute systems, such as the one we consider here). We do not expect this to impact our conclusions, however, as our study is based on identifying where collisions occur, which depends on the relative values of the collision rate in di erent regions of the ow and not on the absolute values. The rate of collisions depends, of course, on the relative velocity of particles at contact [36{ 39]. For tracers (St = 0) or nearly-tracer particles (St 0), this is determined by uid velocity gradients (/  ), which increase in magnitude as the ow becomes more turbulent. This picture, underlying the work of Sa man and Turner [40], is blind to ow structures: It disregards whether the local velocity gradient arises from rotation or strain. Inertial particles (St > 0), e.g., droplets in air, preferentially concentrate, thereby increasing their local number density [21]. They can also attain relative velocities much larger than that of the underlying ow. Dubbed the sling e ect [41], these events correspond to the formation of singularities or caustics in the particle velocity eld [42, 43]. Although clustering and caustics have been tied to the centrifugal ejection of heavy particles out of vortices, they also occur in smooth random ows that are devoid of structure [44{48]. Consequently, the presence of these e ects does not necessarily imply that collisions sense the structures of turbulence. To unambiguously determine the in uence of the local ow eld, we must begin with tracers which remain uniformly distributed in space. To allow for collisions, the radii of these particles are kept (arti cially) nite, while their inertia is ignored. According to the Sa man-Turner theory [40], collisions should occur uniformly between any two regions that possess the same velocity gradient magnitude, regardless of whether these regions are vortical or straining. We now examine this 4 (a) 3 (b) -0.5 -1.0 1 1 -1.5 0 0 0 1 2 3 0 1 2 3 2 2 FIG. 2. Bi-probability distribution functions P (R ; S ) for inertia-less (tracer) particles, corresponding to 2 2 the values of R and S sampled by (a) particles and (b) collision locations, which show the disproportionate bias towards collisions in strain-dominated regions. (c) (b) 2 ● 3 ●● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● 1.5 ● ● ● ● ● ● ● ● ● ● ● ● ● 2 ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ●● ●● ● 1 ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● 1 ● ● ● ● ● ● ● ● 0.5 ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●●● ● ●● ● ● ●● ●● ●● ● ● ●●●● ● ● ● ● ●● ● ● ●●● ● ●● ●● ●●● ●● ● ● ● ● ● ●●●●● ● ●●●● ●● ●● 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 2 2 FIG. 3. (a) Contours corresponding to P (R ; S ) = 0:02; 0:1, sampled by inertia-less particles (dashed) and their collisions (solid). The blue and red shaded portions correspond to Q < 0:3 and Q > +0:3, and indicate regions dominated by strain and vorticity, respectively. Conditional probability distributions of (b) the cosine of the angle of collision , for data corresponding to the shaded portions of panel (a), and (c) the approach velocity at collision v , normalized by the Kolmogorov velocity v . These panels clearly jj illustrate why straining regions are more e ective at generating collisions. hypothesis, bearing in mind that a discrepancy will implicate ow structures that are intrinsically more e ective at causing collisions. 2 2 Towards this end, we calculate the values of R and S along particle trajectories, as well as at collision locations. The results for inertia-less particles are presented as joint probability 2 2 distributions functions P (R ; S ) in Fig. 2a and Fig. 2b, respectively. It is immediately clear that 2 2 2 2 collisions under-sample vortical regions (R > S ) and over-sample straining regions (S > R ), 2 2 relative to where particles reside. This is also seen in Fig. 3a, which overlays contours of P (R ; S ) = 0:1 and 0.02, for particles (dashed) and collisions (solid). The strain (vorticity) dominated portion of this plot is shaded in blue (red), and corresponds to Q < 0:3 (Q > +0:3). This surprising result is an outcome of the distinct ow topologies of these regions which cause particles to approach each other di erently. Fig. 3b presents the distribution of the cosine of the collision angle (), for straining (blue) and vortical (red) regions.  is de ned as the angle between the relative velocity vector (V V ) and the separation vector (X X ) at collision. Particles p1 p2 p1 p2 5 1.2 0.0 ● ● ● ● ● (a) 5 ● ● (b) ● (c) ● ● ● ● ● 3 ● ● ● ● -0.1 ● ● ● ● ● ● ● ● ● 1 ● ● ● -0.2 0.8 0.3 -0.3 ● ● ● ● ● ● ● ● ● -0.4 0.6 0.1 0.01 0.1 0.3 1 10 0.01 0.1 0.3 1 10 0.01 0.1 0.3 1 10 FIG. 4. (a) Average Q, sampled by particles (red-dashed) and collisions (black-solid), as a function of St . (b) Average particle number density in vortical and straining regions (positive and negative Q), plotted as a function of St , and normalized by the domain-average number density n  . (c) Average approach velocity for collisions in vortical and straining regions, normalized by v in straining regions tend to collide in a head-on or rear-end manner (  0 or ). In either case, a large fraction of the velocity di erence between particles contributes to their rate of approach or collision velocity (v ). On the other hand, particles in vortices undergo collisions that are closer jj to being side-on, in which case the separation vector is nearly perpendicular to the relative velocity. This results in lower approach velocities in vortical regions, as shown in Fig. 3c. Consequently, over a given time interval, fewer particles will collide in vortical regions compared to straining regions with the same magnitude of the velocity gradient and particle number density. How does particle inertia a ect this picture? Figure 4a presents the average value of Q sampled by particles (dashed-red) and their collision locations (solid-black) as a function of St . At St = 0, hQi is 0 for particles and -0.04 for collisions. Remarkably, this o set is strongly ampli ed by inertia and reaches a maximum around St  0:3, beyond which particles begin to de-correlate from the underlying ow and eventually collide uniformly. The preference of inertial particles (St > 0) to collide in straining regions has been reported previously by Perrin and Jonker [49, 50]. Our results demonstrate that this e ect is not fundamentally tied to particle inertia, but rather is an ampli cation of a di erence that exists even for inertia-less tracers, raising the question: How does particle inertia selectively enhance collisions in straining regions? One possible explanation is provided by preferential concentration: Heavy particles are cen- trifuged out of rotational regions, and thus tend to accumulate in straining zones just outside vortices [43]. This causes the number density to increase in straining regions, at the expense of vortical zones, as shown in Fig. 4b. Here, n is a coarse-grained number density, obtained by di- viding the domain into bins of size 20. The average Q in each bin is used to distinguish between vortical (Q > 0) and straining regions (Q < 0) and obtain the conditionally averaged number density. All else being equal, higher number densities imply larger collision rates [20]. However, we see that the maximum di erence in number densities occurs near St  1, which is not where the maximum di erence in hQi is seen (Fig. 4a). Hence another mechanism must be involved. Inertia is also known to increase the relative velocity between neighboring particles [36{38], which should result in higher collision velocities. On examining this e ect in straining and vortical regions separately, we nd that it is stronger in straining regions and, in fact, has no impact on vortical regions for small St . This is demonstrated in Fig. 4c, which presents the average values of v , conditioned on Q. It appears that head-on (or rear-end) collisions, which prevail in jj straining zones, are more amenable to being sped-up by inertia than side-on collisions. Notably, the 6 -0.25 0.35 (b) (a) 2.0 (c) -0.30 0.30 0.1 1.5 -0.35 0.6 0.25 1 1.0 -0.40 0.20 -0.45 0.5 0.15 -0.50 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 FIG. 5. Representative plots of (a) the maximum value of Q sampled by the particle inside the vortical region; (b) the average value of Q at collision; and (c) the corresponding approach velocity at collisions obtained from Lagrangian tracking of particles that collide in straining regions (Q < 0) after leaving a vortical region and conditionally averaged on the time taken to collide after entering the straining region (t ). These plots clearly illustrate the singular importance of vortex-strain worm-rolls as discussed in strain the text. maximum di erence between collision velocities occurs near St  0:3, which matches well with the maximum di erence in hQi (Fig. 4a). Thus, larger approach rates, rather than number densities, appear to be the primary reason for the e ectiveness of straining regions in creating collisions. Thus far, we have considered vortical and straining regions individually. Particle inertia, how- ever, permits structures within a distance of  V to in uence a collision. This raises the possibility p p of vortical and straining regions conspiring to enhance collisions, especially for moderately large St . For example, the geometry of vortex-strain worm-rolls (Fig. 1) will cause particles in intense vortex tubes to be rapidly ejected into strong straining sheets, where they are very likely to collide. We search for evidence of this e ect by tracing, backward in time, all particles that collide in straining regions. For the subset that do have at least one particle originating from a vortical region (Q > 0), we record (i) the time taken to collide after leaving the vortical region and entering the straining zone (t ), as indicated by Q changing sign; (ii) the strength of the vortex, measured strain in terms of the maximum positive value of Q sampled inside the vortical region (Q ); (iii) vortex the intensity of straining at the collision location (Q ); and (iv) the collision velocity. While col measuring the vortex strength, we only back-track for a time of 3 within the vortical region, to ensure that the Q values obtained are relevant to the subsequent collision. The conclusions are insensitive to the exact value of this threshold. To facilitate back-tracking, we store the values of Q along all particle trajectories, at time intervals of  =6, for the entire duration of the simulations. This avoids having to store the velocity eld at each time step, for integrating the particle motion backward in time (as done in [51]), which, given the rarity of collisions and the consequent long simulation time, would require prohibitively large amounts of computer storage. Figure 5 presents the results of this backward-in-time Lagrangian calculation, conditionally averaged on the time taken to collide after leaving a vortical region, t , for St = 1; 0:6 and strain 0:1. The data for moderately large St (1 and 0.6), reveal the impact of vortex-strain worm-rolls. Particles that collide quickly (small t ), are found to originate from more intense vortical strain regions (Fig. 5a) and to collide in stronger straining regions (Fig. 5b). This signature weakens considerably for less inertial (St = 0:1) particles which are mildly ejected and relax faster to the local straining ow. The collision velocities for small t are also systematically larger (Fig. 5c). The standard strain deviation about each data point (not shown for clarity) is of the order of the average value. Thus, 7 several small t collisions have very large collision velocities, indicative of caustics/sling events, strain which are known to dominate the collision rate for St > 0:5 [20, 24]. Traditionally, these have been linked to rapid ejection from vortices [41]. Our results indicate that this is only half the story: Straining sheets which envelope strong vortices also contribute to generating violent collisions and enhancing collision rates. All this leads one to ask how collisions are a ected when structures change. The in uence of Re is particularly important to consider, as the estimated values for natural ows are orders of magnitude larger than what can be attained in simulations [17, 18]. Increasing Re results in higher intermittency, which translates into more intense structures (see, e.g., Ref. [52] for a similar study of a model stretched-vortex), but which occupy smaller volumes. These competing e ects are known to produce a non-monotonic variation of particle clustering [53]. For collisions in particular, we have checked explicitly through simulation that as we increase the Reynolds numbers from Re = 69 to 196, unsurprisingly, the di erences between vortical and straining regions magni es; therefore our conclusions, substantiated in this paper from simulations with Re = 196 hold. Nevertheless, we should keep in mind that Re = 196 is still a reasonably modest Reynolds number and far from those seen in atmospheric conditions. Therefore, it remains to be seen in a systematic way how this phenomenon is a ected in a higher Reynolds number ow. However, given our present understanding of the e ect of increasing Reynolds numbers on turbulent structures, we expect that the central results of our work will remain unchanged and the e ect that we elucidate will only show up more clearly. Flow structures can also be signi cantly modi ed by new physical interactions, for example, condensation of water vapour on cloud droplets, which releases latent heat and energizes small scales [54], and elastic feedback from polymers that suppresses small-scale motions [55, 56]. Studying collisions in these complex ows is left for future work. Furthermore, in order to clearly examine the e ect reported in this work, we have neglected the role of gravity. We know that in the limit of small Froude numbers (ratio of turbulent to gravitational acceleration) [57], heavy droplets settle in a way where the uid structures are sampled di erently from the case when Froude numbers are large. In such low Froude number ows, it remains to be seen how the mechanism described in this paper is modi ed. Before we conclude, it is essential to place our work in the context of turbulent transport problems|a canonical example being that of rain-initiation in warm clouds|which have applica- tion across the areas of non-equilibrium statistical physics, geophysics, oceanography, astrophysics and atmospheric sciences. Understanding these problems demands not only an appreciation of how fast droplets sediment, collide and coalesce (in which tremendous advances have been made in recent years) but also knowledge of where such processes are most likely to occur. This paper, therefore, contributes to a fuller understanding of this question. ACKNOWLEDGMENTS We thank J er emie Bec, S. Ravichandran and Siddhartha Mukherjee for useful suggestions and discussions, which were facilitated in part by a program organized at ICTS: Turbulence from Angstroms to Light Years (ICTS/Prog-taly2018/01). The simulations were performed on the ICTS clusters Mowgli and Mario as well as the work stations from the project ECR/2015/000361: Goopy 8 and Bagha. SSR acknowledges DST (India) project ECR/2015/000361 for nancial support. [1] U. Frisch, Turbulence: The legacy of A N Kolmogorov (Cambridge University Press, Cambridge, UK, 1996). [2] A. Tsinober, An informal conceptual introduction to turbulence (Springer, New York, USA, 2009). [3] J. Schumacher, R. M. Kerr, and K. Horiuti, \Structure and dynamics of vorticity in turbulence," in Ten Chapters in Turbulence, edited by P. A. Davidson, Y. Kaneda, and K. R. Sreenivasan (Cambridge University Press, 2012) p. 4386. [4] U. Frisch, A. Pomyalov, I. Procaccia, and S. S. Ray, \Turbulence in noninteger dimensions by fractal fourier decimation," Phys. Rev. Lett. 108, 074501 (2012). [5] M. Buzzicotti, A. Bhatnagar, L. Biferale, A. S. Lanotte, and S. S. Ray, \Lagrangian statistics for navier-stokes turbulence under fourier-mode reduction: fractal and homogeneous decimations," New J. Phys. 18, 113047 (2016). [6] A. S. Lanotte, S. K. Malapaka, and L. Biferale, \On the vortex dynamics in fractal fourier turbulence," Eur. Phys. J. E 39, 49 (2016). [7] B. W. Ze , D. D. Lanterman, R. McAllister, R. Roy, E. J. Kostelich, and D. P. Lathrop, \Measuring intense rotation and dissipation in turbulent ows," Nature 34, B479{B498 (2003). [8] R. Pandit, P. Perlekar, and S. S. Ray, \Statistical properties of turbulence: An overview," Pramana- Journal of Physics 73, 157 (2009). [9] J. Schumacher, B. Eckhardt, and C. R. Doering, \Extreme vorticity growth in navier-stokes turbu- lence," Phys. Lett. A 374, 861{865 (2010). [10] P. Gotzfried, B. Kumar, R. A. Shaw, and J. Schumacher, \Droplet dynamics and ne-scale structure in a shearless turbulent mixing layer with phase changes," J. Fluid Mech. 814, 452483 (2017). [11] M. Guala, A. Liberzon, A. Tsinober, and W. Kinzelbach, \An experimental investigation on lagrangian correlations of small-scale turbulence at low reynolds number," J. Fluid Mech. 574, 405427 (2007). [12] P. E. Hamlington, J. Schumacher, and W. J. A. Dahm, \Local and nonlocal strain rate elds and vorticity alignment in turbulent ows," Phys. Rev. E 77, 026303 (2008). [13] S. S. Ray, \Thermalised solutions, statistical mechanics and turbulence: An overview of some recent results," Pramana-Journal of Physics 84, 395 (2015). [14] M. Wilczek, \New insights into the ne-scale structure of turbulence," J. Fluid Mech. 784, 14 (2015). [15] J. M. Lawson and J. R. Dawson, \On velocity gradient dynamics and turbulent structure," J. Fluid Mech. 780, 6098 (2015). [16] S. S. Ray, \Non-intermittent turbulence: Lagrangian chaos and irreversibility," Phys. Rev. Fluids (Rapid) 18, 072601(R) (2018). [17] J. J. Lissauer, \Planet formation," Annu. Rev. Astron. and Astrophys. 31, 129{172 (1993). [18] W. W. Grabowski and L.-P. Wang, \Growth of cloud droplets in a turbulent environment," Annu. Rev. Fluid Mech. 45, 293{324 (2013). [19] S. Chen, M.-K. Yau, P. Bartello, and L. Xue, \Bridging the condensation{collision size gap: a direct numerical simulation of continuous droplet growth in turbulent clouds," Atmos. Chem. Phys. 18, 7251{ 7262 (2018). [20] A. Pumir and M. Wilkinson, \Collisional aggregation due to turbulence," Annu. Rev. Conden. Ma. P. 7, 141{170 (2016). [21] S. Sundaram and L. R. Collins, \Collision statistics in an isotropic particle-laden turbulent suspension. part 1. direct numerical simulations," J. Fluid Mech. 335, 75109 (1997). [22] J. Bec, S. S. Ray, E. W. Saw, and H. Homann, \Abrupt growth of large aggregates by correlated coalescences in turbulent ow," Phys. Rev. E 93, 031102(R) (2016). [23] R. Onishi, K. Matsuda, and K. Takahashi, \Lagrangian tracking simulation of droplet growth in turbulenceturbulence enhancement of autoconversion rate," J. Atmos. Sci. 72, 2591{2607 (2015). [24] M. Vokuhle, A. Pumir, E. L ev^ eque, and M. Wilkinson, \Prevalence of the sling e ect for enhancing collision rates in turbulent suspensions," J. Fluid Mech. 749, 841852 (2014). [25] P. J. Ireland, A. D. Bragg, and L. R. Collins, \The e ect of reynolds number on inertial particle dynamics in isotropic turbulence. part 1. simulations without gravitational e ects," J. Fluid Mech. 9 796, 617658 (2016). [26] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, Spectral methods: fundamental in single domains (Springer-Verlag, Berlin, Germany, 2006). [27] Y. Dubief and F. Delcayre, \On coherent-vortex identi cation in turbulence," J. Turbul. 1, N11 (2000). [28] H. M. Blackburn, N. N. Mansour, and B. J. Cantwell, \Topology of ne-scale motions in turbulent channel ow," J. Fluid Mech. 310, 269292 (1996). [29] M. S. Chong, A. E. Perry, and B. J. Cantwell, \A general classi cation of threedimensional ow elds," Phys. Fluids A 2, 765{777 (1990). [30] S. Ravichandran, P. Deepu, and R. Govindarajan, \Clustering of heavy particles in vortical ows: a selective review," S adhan a 42, 597{605 (2017). [31] V. Armenio and V. Fiorotto, \The importance of the forces acting on particles in turbulent ows," Phys. Fluids 13, 2437{2440 (2001). [32] M. A. T. van Hinsberg, H. J. H. Clercx, and F. Toschi, \Enhanced settling of nonheavy inertial particles in homogeneous isotropic turbulence: The role of the pressure gradient and the basset history force," Phys. Rev. E 95, 023106 (2017). [33] M. van Aartrijk and H. J. H. Clercx, \Vertical dispersion of light inertial particles in stably strati ed turbulence: The in uence of the basset force," Phys. Fluids 22, 013301 (2010). [34] P. J. Ireland, T. Vaithianathan, P. S. Sukheswalla, B. Ray, and L. R. Collins, \Highly parallel particle- laden ow solver for turbulence research," Comput. Fluids 76, 170 { 177 (2013). [35] M. A. T. van Hinsberg, J. H. M. T. Boonkkamp, F. Toschi, and H. J. H. Clercx, \On the eciency and accuracy of interpolation methods for spectral codes," SIAM J. Sci. Comput. 34, B479{B498 (2012). [36] E.-W. Saw, G. P. Bewley, E. Bodenschatz, S. S. Ray, and J. Bec, \Extreme uctuations of the relative velocities between droplets in turbulent air ow," Phys. Fluids 26, 111702 (2014). [37] M. James and S. S. Ray, \Enhanced droplet collision rates and impact velocities in turbulent ows: The e ect of poly-dispersity and transient phases," Sci. Rep. 7, 12231 (2017). [38] A. Bhatnagar, K. Gustavsson, and D. Mitra, \Statistics of the relative velocity of particles in turbulent ows: Monodisperse particles," Phys. Rev. E 97, 023105 (2018). [39] A. Bhatnagar, K. Gustavsson, B. Mehlig, and D. Mitra, \Relative velocities in bidisperse turbulent aerosols: Simulations and theory," Phys. Rev. E 98, 063107 (2018). [40] P. G. Sa man and J. S. Turner, \On the collision of drops in turbulent clouds," J. Fluid Mech. 1, 1630 (1956). [41] G. Falkovich, A. Fouxon, and M. G. Stepanov, \Acceleration of rain initiation by cloud turbulence," Nature 419, 151{154 (2002). [42] M. Wilkinson, B. Mehlig, and V. Bezuglyy, \Caustic activation of rain showers," Phys. Rev. Lett. 97, 048501 (2006). [43] S. Ravichandran and R. Govindarajan, \Caustics and clustering in the vicinity of a vortex," Phys. Fluids 27, 033305 (2015). [44] J. Bec, A. Celani, M. Cencini, and S. Musacchio, \Clustering and collisions of heavy particles in random smooth ows," Phys. Fluids 17, 073301 (2005). [45] M. Wilkinson, B. Mehlig, S. Ostlund, and K. P. Duncan, \Unmixing in random ows," Phys Fluids 19, 113303 (2007). [46] J. Bec, S. Musacchio, and S. S. Ray, \Sticky elastic collisions," Phys. Rev. E 87, 063013 (2013). [47] K. Gustavsson and B. Mehlig, \Distribution of velocity gradients and rate of caustic formation in turbulent aerosols at nite kubo numbers," Phys. Rev. E 87, 023016 (2013). [48] K. Gustavsson and B. Mehlig, \Relative velocities of inertial particles in turbulent aerosols," J. Turbul. 15, 34{69 (2014). [49] V. E. Perrin and H. J. J. Jonker, \Preferred location of droplet collisions in turbulent ows," Phys. Rev. E 89, 033005 (2014). [50] V. E. Perrin and H. J. J. Jonker, \E ect of the eigenvalues of the velocity gradient tensor on particle collisions," J. Fluid Mech. 792, 3649 (2016). [51] M. Wan, Z. Xiao, C. Meneveau, G. L. Eyink, and S. Chen, \Dissipation-energy ux correlations as evidence for the lagrangian energy cascade in turbulence," Phys. Fluids 22, 061702 (2010). [52] L. Agasthya, J. R. Picardo, S. Ravichandran, R. Govindarajan, and S. S. Ray, \Droplet collisions in turbulence: Insights from a burgers vortex," arXiv e-prints (2018), arXiv:1811.04486. [53] R. Onishi and J. C. Vassilicos, \Collision statistics of inertial particles in two-dimensional homogeneous 10 isotropic turbulence with an inverse cascade," J. Fluid Mech. 745, 279299 (2014). [54] S. Ravichandran and R. Govindarajan, \Vortex-dipole collapse induced by droplet inertia and phase change," J. Fluid Mech. 832, 745776 (2017). [55] A. Liberzon, M. Guala, W. Kinzelbach, and A. Tsinober, \On turbulent kinetic energy production and dissipation in dilute polymer solutions," Phys. Fluids 18, 125101 (2006). [56] P. Perlekar, D. Mitra, and R. Pandit, \Direct numerical simulations of statistically steady, homoge- neous, isotropic uid turbulence with polymer additives," Phys. Rev. E 82, 066313 (2010). [57] J. Bec, H. Homann, and S. S. Ray, \Gravity-driven enhancement of heavy particle clustering in turbulent ow," Phys. Rev. Lett. 112, 184501 (2014). http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nonlinear Sciences arXiv (Cornell University)

Flow structures govern particle collisions in turbulence

Download PDF
10 pages

Loading next page...
 
/lp/arxiv-cornell-university/flow-structures-govern-particle-collisions-in-turbulence-J02qX9kroy

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

ISSN
2469-990X
eISSN
ARCH-3338
DOI
10.1103/PhysRevFluids.4.032601
Publisher site
See Article on Publisher Site

Abstract

1, 2,y 1,z 1,x Jason R. Picardo, Lokahith Agasthya, Rama Govindarajan, and Samriddhi Sankar Ray International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bangalore 560089, India Indian Institute for Science Education and Research, Pune 411008, India The role of the spatial structure of a turbulent ow in enhancing particle collision rates in suspensions is an open question. We show and quantify, as a function of particle inertia, the correlation between the multiscale structures of turbulence and particle collisions: Straining zones contribute predominantly to rapid head-on collisions compared to vortical regions. We also discover the importance of vortex-strain worm-rolls, which goes beyond ideas of pref- erential concentration and may explain the rapid growth of aggregates in natural processes, such as the initiation of rain in warm clouds. Turbulence is riddled with a hierarchy of interacting vortical and straining structures (Fig. 1), which are closely related to its characteristic intermittent and non-Gaussian statistics [1{6]. The most intense structures typically occur near each other, in the form of vortex tubes surrounded by straining sheets, as shown in Fig. 1. This organization|a sort of vortex-strain worm-rolls |is characteristic of turbulent ows [7{10], and its origin and dynamical implications continue to be investigated [11{16]. These structures distinguish fully developed turbulence from purely random ow elds, and must play an important role in many aspects of turbulent transport. The most important of these|because it remains central to our understanding of phenomena as diverse as the formation of planets in circumstellar disks [17] or the initiation of rain in warm clouds [18, 19]| is the growth of macroscopic aggregates, due to collisions and coalescences, from nuclei-particles (dust or aerosols) suspended in a turbulent ow. The role of the underlying turbulent carrier ow is critical: Estimates of, e.g., the rate of growth of these aggregates in the absence of such ows do not agree with that seen in nature [20]. Indeed, the explanation of such rapid growth through coalescence, demonstrated [21{23] and quanti ed in terms of ow statistics [20, 24, 25], is rooted in the ability of turbulent ows to enhance the rate of collisions between nuclei seed-particles. A critical discovery, due to Bec, et al. [22], was to nd the precise connection between the inter- mittent (multiscaling) nature of the carrier turbulent ow and the accelerated growth of aggregates. And yet the implied correlation between the structure of the ow and droplet collisions-coalescences remains unknown. Indeed, there is evidence to suggest that ow structures matter. But the ques- tion is how and when? In this paper we answer this question comprehensively, based on direct numerical simulations (DNSs), and show how straining regions are intrinsically more e ective at generating collisions than vortical ones even for uniformly distributed inertia-less particles. Particle inertia widens this discrepancy, not simply by preferential concentration, but also by selectively increasing the collision velocities in straining zones. This is because straining regions have a larger proportion of head-on or rear-end collisions, as opposed to side-on collisions, which are predominant in vortical regions. Consequently, a larger fraction of the velocity gradient in straining zones is translated into the particle approach velocity. Finally, and most strikingly, we show how intense vorticity and strain, cohabiting as vortex-strain worm-rolls, conspire to generate rapid, violent collisions. We therefore consider an incompressible (ru = 0) turbulent ow whose velocity u is a solution to the Navier-Stokes equation @ u + (ur)u = rp + r u + f (1) jrpicardo@icts.res.in; picardo21@gmail.com lokahith.agasthya@students.iiserpune.ac.in rama@icts.res.in samriddhisankarray@gmail.com arXiv:1810.10285v2 [physics.flu-dyn] 26 Mar 2019 2 FIG. 1. A representative snapshot of three-dimensional contours of Q showing intense vortex tubes (opaque p p 2 2 red: +5:6 hQ i) enveloped by dissipative, straining sheets (transparent blue: 2 hQ i). 3 3 where  is the kinematic viscosity. We perform DNSs on a tri-periodic domain with N = 512 grid points, by using a standard de-aliased pseudo-spectral solver [26] and a second-order Adams- Bashford scheme for time-integration. A statistically stationary, homogeneous and isotropic ow is maintained by the time-dependent large-scale forcing f , which injects a constant amount of energy, and hence dissipation , into the rst two wavenumber shells. The Kolmogorov length 3 1=4 = ( =) satis es k  1:7 (where k = 2N=3 is the maximum resolved wavenumber). max max The Taylor-Reynolds number Re = 2E 5=(3) = 196, where E is the total kinetic energy. The compensated energy spectrum from such a simulation shows an inertial range which is certainly less than a decade but still resolved (see, e.g., Fig. 1 (a) for N = 512 and D = 3:00 in Ref. [5]). To identify vortical and straining structures in the ow, we use the second invariant of the 2 2 velocity gradient tensor Q = (R S )=2 [27{29], de ned through the velocity gradient tensor 1=2 A =  ru (normalized by the Kolmogorov time  = (=) ) which yields the symmetric strain T T rate tensor S = (A +A )=2 and the anti-symmetric rotation rate tensor R = (AA )=2. Regions where Q > 0 (Q < 0) are dominated by vorticity (irrotational strain) [25]. Figure 1 presents contours of large positive and negative values of Q, which reveal characteristic vortex-strain worm- rolls. We introduce in the ow 10 identical particles, each having a sub-Kolmogorov radius a = =3 and a density  much larger than that of the carrier- uid  . The particles occupy a small volume p f fraction of O(10 ) and their in uence on the ow is negligible. Since the Reynolds number associated with their slip velocities is small, and    , the evolution of particle trajectories p f X (t) is determined by the simpli ed Maxey-Riley equations [30{33]: dX dV 1 p p = V ; = [V u(X ; t)] (2) p p p dt dt p 3 where  = 2a  =(9 ), the particle relaxation time, yields the Stokes number (St =  = ), p p f p which provides a non-dimensional measure of the particle's inertia. We consider several families of particles, with St ranging from 0 to 16.75 and use an exponential integration scheme [34]. The case of tracers St = 0 is handled separately by using a second order Runge-Kutta time-stepper. The uid velocity at the particle position is obtained via fourth-order B-spline interpolation [35]. It is important to note that the particle radius in uences particle transport in two distinct ways: It sets the value of St, while also determining the separation between particle centers at collision. As we are considering heavy particles ( = > 1), the radius must be very small for small values of St. This makes the gathering of collision statistics in this dilute suspension increasingly impractical as St ! 0, unless we consider an arti cially enlarged collision radius. Indeed, for tracers, by de nition, any choice of radius is arbitrary. Our choice of =3 is physically consistent for particles with St > 0:1. However, for smaller St, this particle radius would imply  = < 1 p f which contradicts our heavy particle assumption. Indeed, to accurately describe particle motion, the simpli ed Maxey-Riley equations (2) require  = > O(10 ) [31{33]. Nevertheless, it is p f convenient to consider this enlarged collision radius, solely for the purpose of detecting collisions. As a check, we actually performed simulations with particle radius =10 (where the problem is mitigated) and found our results unchanged. This can be rationalised by noting that particles move ballistically in the small distance which separates the =3 and the more realistic =10 radii and hence make our collision results insensitive to the precise choice of the radius. Nevertheless, we do report results for the larger radius because the collision statistics are much poorer with the smaller radius. After the randomly-seeded particles have settled into a statistically stationary distribution, we begin detecting collisions using an algorithm similar to that in Ref. [21]. After a collision is detected, the two particles are allowed to move past each other without any modi cation to their trajectories. This ghost collision approach is a standard approximation that, by ignoring coalescence, allows one to measure collision rates while the particle distribution remains in a statistically stationary state. This feature allows us to identify and compare the regions of the ow where most of the collisions occur to those regions where most of the particles reside. Of course, because particles never coalesce, the collision rates are over-predicted (this issue is less severe for dilute systems, such as the one we consider here). We do not expect this to impact our conclusions, however, as our study is based on identifying where collisions occur, which depends on the relative values of the collision rate in di erent regions of the ow and not on the absolute values. The rate of collisions depends, of course, on the relative velocity of particles at contact [36{ 39]. For tracers (St = 0) or nearly-tracer particles (St 0), this is determined by uid velocity gradients (/  ), which increase in magnitude as the ow becomes more turbulent. This picture, underlying the work of Sa man and Turner [40], is blind to ow structures: It disregards whether the local velocity gradient arises from rotation or strain. Inertial particles (St > 0), e.g., droplets in air, preferentially concentrate, thereby increasing their local number density [21]. They can also attain relative velocities much larger than that of the underlying ow. Dubbed the sling e ect [41], these events correspond to the formation of singularities or caustics in the particle velocity eld [42, 43]. Although clustering and caustics have been tied to the centrifugal ejection of heavy particles out of vortices, they also occur in smooth random ows that are devoid of structure [44{48]. Consequently, the presence of these e ects does not necessarily imply that collisions sense the structures of turbulence. To unambiguously determine the in uence of the local ow eld, we must begin with tracers which remain uniformly distributed in space. To allow for collisions, the radii of these particles are kept (arti cially) nite, while their inertia is ignored. According to the Sa man-Turner theory [40], collisions should occur uniformly between any two regions that possess the same velocity gradient magnitude, regardless of whether these regions are vortical or straining. We now examine this 4 (a) 3 (b) -0.5 -1.0 1 1 -1.5 0 0 0 1 2 3 0 1 2 3 2 2 FIG. 2. Bi-probability distribution functions P (R ; S ) for inertia-less (tracer) particles, corresponding to 2 2 the values of R and S sampled by (a) particles and (b) collision locations, which show the disproportionate bias towards collisions in strain-dominated regions. (c) (b) 2 ● 3 ●● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● 1.5 ● ● ● ● ● ● ● ● ● ● ● ● ● 2 ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ●● ●● ● 1 ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● 1 ● ● ● ● ● ● ● ● 0.5 ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●●● ● ●● ● ● ●● ●● ●● ● ● ●●●● ● ● ● ● ●● ● ● ●●● ● ●● ●● ●●● ●● ● ● ● ● ● ●●●●● ● ●●●● ●● ●● 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 2 2 FIG. 3. (a) Contours corresponding to P (R ; S ) = 0:02; 0:1, sampled by inertia-less particles (dashed) and their collisions (solid). The blue and red shaded portions correspond to Q < 0:3 and Q > +0:3, and indicate regions dominated by strain and vorticity, respectively. Conditional probability distributions of (b) the cosine of the angle of collision , for data corresponding to the shaded portions of panel (a), and (c) the approach velocity at collision v , normalized by the Kolmogorov velocity v . These panels clearly jj illustrate why straining regions are more e ective at generating collisions. hypothesis, bearing in mind that a discrepancy will implicate ow structures that are intrinsically more e ective at causing collisions. 2 2 Towards this end, we calculate the values of R and S along particle trajectories, as well as at collision locations. The results for inertia-less particles are presented as joint probability 2 2 distributions functions P (R ; S ) in Fig. 2a and Fig. 2b, respectively. It is immediately clear that 2 2 2 2 collisions under-sample vortical regions (R > S ) and over-sample straining regions (S > R ), 2 2 relative to where particles reside. This is also seen in Fig. 3a, which overlays contours of P (R ; S ) = 0:1 and 0.02, for particles (dashed) and collisions (solid). The strain (vorticity) dominated portion of this plot is shaded in blue (red), and corresponds to Q < 0:3 (Q > +0:3). This surprising result is an outcome of the distinct ow topologies of these regions which cause particles to approach each other di erently. Fig. 3b presents the distribution of the cosine of the collision angle (), for straining (blue) and vortical (red) regions.  is de ned as the angle between the relative velocity vector (V V ) and the separation vector (X X ) at collision. Particles p1 p2 p1 p2 5 1.2 0.0 ● ● ● ● ● (a) 5 ● ● (b) ● (c) ● ● ● ● ● 3 ● ● ● ● -0.1 ● ● ● ● ● ● ● ● ● 1 ● ● ● -0.2 0.8 0.3 -0.3 ● ● ● ● ● ● ● ● ● -0.4 0.6 0.1 0.01 0.1 0.3 1 10 0.01 0.1 0.3 1 10 0.01 0.1 0.3 1 10 FIG. 4. (a) Average Q, sampled by particles (red-dashed) and collisions (black-solid), as a function of St . (b) Average particle number density in vortical and straining regions (positive and negative Q), plotted as a function of St , and normalized by the domain-average number density n  . (c) Average approach velocity for collisions in vortical and straining regions, normalized by v in straining regions tend to collide in a head-on or rear-end manner (  0 or ). In either case, a large fraction of the velocity di erence between particles contributes to their rate of approach or collision velocity (v ). On the other hand, particles in vortices undergo collisions that are closer jj to being side-on, in which case the separation vector is nearly perpendicular to the relative velocity. This results in lower approach velocities in vortical regions, as shown in Fig. 3c. Consequently, over a given time interval, fewer particles will collide in vortical regions compared to straining regions with the same magnitude of the velocity gradient and particle number density. How does particle inertia a ect this picture? Figure 4a presents the average value of Q sampled by particles (dashed-red) and their collision locations (solid-black) as a function of St . At St = 0, hQi is 0 for particles and -0.04 for collisions. Remarkably, this o set is strongly ampli ed by inertia and reaches a maximum around St  0:3, beyond which particles begin to de-correlate from the underlying ow and eventually collide uniformly. The preference of inertial particles (St > 0) to collide in straining regions has been reported previously by Perrin and Jonker [49, 50]. Our results demonstrate that this e ect is not fundamentally tied to particle inertia, but rather is an ampli cation of a di erence that exists even for inertia-less tracers, raising the question: How does particle inertia selectively enhance collisions in straining regions? One possible explanation is provided by preferential concentration: Heavy particles are cen- trifuged out of rotational regions, and thus tend to accumulate in straining zones just outside vortices [43]. This causes the number density to increase in straining regions, at the expense of vortical zones, as shown in Fig. 4b. Here, n is a coarse-grained number density, obtained by di- viding the domain into bins of size 20. The average Q in each bin is used to distinguish between vortical (Q > 0) and straining regions (Q < 0) and obtain the conditionally averaged number density. All else being equal, higher number densities imply larger collision rates [20]. However, we see that the maximum di erence in number densities occurs near St  1, which is not where the maximum di erence in hQi is seen (Fig. 4a). Hence another mechanism must be involved. Inertia is also known to increase the relative velocity between neighboring particles [36{38], which should result in higher collision velocities. On examining this e ect in straining and vortical regions separately, we nd that it is stronger in straining regions and, in fact, has no impact on vortical regions for small St . This is demonstrated in Fig. 4c, which presents the average values of v , conditioned on Q. It appears that head-on (or rear-end) collisions, which prevail in jj straining zones, are more amenable to being sped-up by inertia than side-on collisions. Notably, the 6 -0.25 0.35 (b) (a) 2.0 (c) -0.30 0.30 0.1 1.5 -0.35 0.6 0.25 1 1.0 -0.40 0.20 -0.45 0.5 0.15 -0.50 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 FIG. 5. Representative plots of (a) the maximum value of Q sampled by the particle inside the vortical region; (b) the average value of Q at collision; and (c) the corresponding approach velocity at collisions obtained from Lagrangian tracking of particles that collide in straining regions (Q < 0) after leaving a vortical region and conditionally averaged on the time taken to collide after entering the straining region (t ). These plots clearly illustrate the singular importance of vortex-strain worm-rolls as discussed in strain the text. maximum di erence between collision velocities occurs near St  0:3, which matches well with the maximum di erence in hQi (Fig. 4a). Thus, larger approach rates, rather than number densities, appear to be the primary reason for the e ectiveness of straining regions in creating collisions. Thus far, we have considered vortical and straining regions individually. Particle inertia, how- ever, permits structures within a distance of  V to in uence a collision. This raises the possibility p p of vortical and straining regions conspiring to enhance collisions, especially for moderately large St . For example, the geometry of vortex-strain worm-rolls (Fig. 1) will cause particles in intense vortex tubes to be rapidly ejected into strong straining sheets, where they are very likely to collide. We search for evidence of this e ect by tracing, backward in time, all particles that collide in straining regions. For the subset that do have at least one particle originating from a vortical region (Q > 0), we record (i) the time taken to collide after leaving the vortical region and entering the straining zone (t ), as indicated by Q changing sign; (ii) the strength of the vortex, measured strain in terms of the maximum positive value of Q sampled inside the vortical region (Q ); (iii) vortex the intensity of straining at the collision location (Q ); and (iv) the collision velocity. While col measuring the vortex strength, we only back-track for a time of 3 within the vortical region, to ensure that the Q values obtained are relevant to the subsequent collision. The conclusions are insensitive to the exact value of this threshold. To facilitate back-tracking, we store the values of Q along all particle trajectories, at time intervals of  =6, for the entire duration of the simulations. This avoids having to store the velocity eld at each time step, for integrating the particle motion backward in time (as done in [51]), which, given the rarity of collisions and the consequent long simulation time, would require prohibitively large amounts of computer storage. Figure 5 presents the results of this backward-in-time Lagrangian calculation, conditionally averaged on the time taken to collide after leaving a vortical region, t , for St = 1; 0:6 and strain 0:1. The data for moderately large St (1 and 0.6), reveal the impact of vortex-strain worm-rolls. Particles that collide quickly (small t ), are found to originate from more intense vortical strain regions (Fig. 5a) and to collide in stronger straining regions (Fig. 5b). This signature weakens considerably for less inertial (St = 0:1) particles which are mildly ejected and relax faster to the local straining ow. The collision velocities for small t are also systematically larger (Fig. 5c). The standard strain deviation about each data point (not shown for clarity) is of the order of the average value. Thus, 7 several small t collisions have very large collision velocities, indicative of caustics/sling events, strain which are known to dominate the collision rate for St > 0:5 [20, 24]. Traditionally, these have been linked to rapid ejection from vortices [41]. Our results indicate that this is only half the story: Straining sheets which envelope strong vortices also contribute to generating violent collisions and enhancing collision rates. All this leads one to ask how collisions are a ected when structures change. The in uence of Re is particularly important to consider, as the estimated values for natural ows are orders of magnitude larger than what can be attained in simulations [17, 18]. Increasing Re results in higher intermittency, which translates into more intense structures (see, e.g., Ref. [52] for a similar study of a model stretched-vortex), but which occupy smaller volumes. These competing e ects are known to produce a non-monotonic variation of particle clustering [53]. For collisions in particular, we have checked explicitly through simulation that as we increase the Reynolds numbers from Re = 69 to 196, unsurprisingly, the di erences between vortical and straining regions magni es; therefore our conclusions, substantiated in this paper from simulations with Re = 196 hold. Nevertheless, we should keep in mind that Re = 196 is still a reasonably modest Reynolds number and far from those seen in atmospheric conditions. Therefore, it remains to be seen in a systematic way how this phenomenon is a ected in a higher Reynolds number ow. However, given our present understanding of the e ect of increasing Reynolds numbers on turbulent structures, we expect that the central results of our work will remain unchanged and the e ect that we elucidate will only show up more clearly. Flow structures can also be signi cantly modi ed by new physical interactions, for example, condensation of water vapour on cloud droplets, which releases latent heat and energizes small scales [54], and elastic feedback from polymers that suppresses small-scale motions [55, 56]. Studying collisions in these complex ows is left for future work. Furthermore, in order to clearly examine the e ect reported in this work, we have neglected the role of gravity. We know that in the limit of small Froude numbers (ratio of turbulent to gravitational acceleration) [57], heavy droplets settle in a way where the uid structures are sampled di erently from the case when Froude numbers are large. In such low Froude number ows, it remains to be seen how the mechanism described in this paper is modi ed. Before we conclude, it is essential to place our work in the context of turbulent transport problems|a canonical example being that of rain-initiation in warm clouds|which have applica- tion across the areas of non-equilibrium statistical physics, geophysics, oceanography, astrophysics and atmospheric sciences. Understanding these problems demands not only an appreciation of how fast droplets sediment, collide and coalesce (in which tremendous advances have been made in recent years) but also knowledge of where such processes are most likely to occur. This paper, therefore, contributes to a fuller understanding of this question. ACKNOWLEDGMENTS We thank J er emie Bec, S. Ravichandran and Siddhartha Mukherjee for useful suggestions and discussions, which were facilitated in part by a program organized at ICTS: Turbulence from Angstroms to Light Years (ICTS/Prog-taly2018/01). The simulations were performed on the ICTS clusters Mowgli and Mario as well as the work stations from the project ECR/2015/000361: Goopy 8 and Bagha. SSR acknowledges DST (India) project ECR/2015/000361 for nancial support. [1] U. Frisch, Turbulence: The legacy of A N Kolmogorov (Cambridge University Press, Cambridge, UK, 1996). [2] A. Tsinober, An informal conceptual introduction to turbulence (Springer, New York, USA, 2009). [3] J. Schumacher, R. M. Kerr, and K. Horiuti, \Structure and dynamics of vorticity in turbulence," in Ten Chapters in Turbulence, edited by P. A. Davidson, Y. Kaneda, and K. R. Sreenivasan (Cambridge University Press, 2012) p. 4386. [4] U. Frisch, A. Pomyalov, I. Procaccia, and S. S. Ray, \Turbulence in noninteger dimensions by fractal fourier decimation," Phys. Rev. Lett. 108, 074501 (2012). [5] M. Buzzicotti, A. Bhatnagar, L. Biferale, A. S. Lanotte, and S. S. Ray, \Lagrangian statistics for navier-stokes turbulence under fourier-mode reduction: fractal and homogeneous decimations," New J. Phys. 18, 113047 (2016). [6] A. S. Lanotte, S. K. Malapaka, and L. Biferale, \On the vortex dynamics in fractal fourier turbulence," Eur. Phys. J. E 39, 49 (2016). [7] B. W. Ze , D. D. Lanterman, R. McAllister, R. Roy, E. J. Kostelich, and D. P. Lathrop, \Measuring intense rotation and dissipation in turbulent ows," Nature 34, B479{B498 (2003). [8] R. Pandit, P. Perlekar, and S. S. Ray, \Statistical properties of turbulence: An overview," Pramana- Journal of Physics 73, 157 (2009). [9] J. Schumacher, B. Eckhardt, and C. R. Doering, \Extreme vorticity growth in navier-stokes turbu- lence," Phys. Lett. A 374, 861{865 (2010). [10] P. Gotzfried, B. Kumar, R. A. Shaw, and J. Schumacher, \Droplet dynamics and ne-scale structure in a shearless turbulent mixing layer with phase changes," J. Fluid Mech. 814, 452483 (2017). [11] M. Guala, A. Liberzon, A. Tsinober, and W. Kinzelbach, \An experimental investigation on lagrangian correlations of small-scale turbulence at low reynolds number," J. Fluid Mech. 574, 405427 (2007). [12] P. E. Hamlington, J. Schumacher, and W. J. A. Dahm, \Local and nonlocal strain rate elds and vorticity alignment in turbulent ows," Phys. Rev. E 77, 026303 (2008). [13] S. S. Ray, \Thermalised solutions, statistical mechanics and turbulence: An overview of some recent results," Pramana-Journal of Physics 84, 395 (2015). [14] M. Wilczek, \New insights into the ne-scale structure of turbulence," J. Fluid Mech. 784, 14 (2015). [15] J. M. Lawson and J. R. Dawson, \On velocity gradient dynamics and turbulent structure," J. Fluid Mech. 780, 6098 (2015). [16] S. S. Ray, \Non-intermittent turbulence: Lagrangian chaos and irreversibility," Phys. Rev. Fluids (Rapid) 18, 072601(R) (2018). [17] J. J. Lissauer, \Planet formation," Annu. Rev. Astron. and Astrophys. 31, 129{172 (1993). [18] W. W. Grabowski and L.-P. Wang, \Growth of cloud droplets in a turbulent environment," Annu. Rev. Fluid Mech. 45, 293{324 (2013). [19] S. Chen, M.-K. Yau, P. Bartello, and L. Xue, \Bridging the condensation{collision size gap: a direct numerical simulation of continuous droplet growth in turbulent clouds," Atmos. Chem. Phys. 18, 7251{ 7262 (2018). [20] A. Pumir and M. Wilkinson, \Collisional aggregation due to turbulence," Annu. Rev. Conden. Ma. P. 7, 141{170 (2016). [21] S. Sundaram and L. R. Collins, \Collision statistics in an isotropic particle-laden turbulent suspension. part 1. direct numerical simulations," J. Fluid Mech. 335, 75109 (1997). [22] J. Bec, S. S. Ray, E. W. Saw, and H. Homann, \Abrupt growth of large aggregates by correlated coalescences in turbulent ow," Phys. Rev. E 93, 031102(R) (2016). [23] R. Onishi, K. Matsuda, and K. Takahashi, \Lagrangian tracking simulation of droplet growth in turbulenceturbulence enhancement of autoconversion rate," J. Atmos. Sci. 72, 2591{2607 (2015). [24] M. Vokuhle, A. Pumir, E. L ev^ eque, and M. Wilkinson, \Prevalence of the sling e ect for enhancing collision rates in turbulent suspensions," J. Fluid Mech. 749, 841852 (2014). [25] P. J. Ireland, A. D. Bragg, and L. R. Collins, \The e ect of reynolds number on inertial particle dynamics in isotropic turbulence. part 1. simulations without gravitational e ects," J. Fluid Mech. 9 796, 617658 (2016). [26] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, Spectral methods: fundamental in single domains (Springer-Verlag, Berlin, Germany, 2006). [27] Y. Dubief and F. Delcayre, \On coherent-vortex identi cation in turbulence," J. Turbul. 1, N11 (2000). [28] H. M. Blackburn, N. N. Mansour, and B. J. Cantwell, \Topology of ne-scale motions in turbulent channel ow," J. Fluid Mech. 310, 269292 (1996). [29] M. S. Chong, A. E. Perry, and B. J. Cantwell, \A general classi cation of threedimensional ow elds," Phys. Fluids A 2, 765{777 (1990). [30] S. Ravichandran, P. Deepu, and R. Govindarajan, \Clustering of heavy particles in vortical ows: a selective review," S adhan a 42, 597{605 (2017). [31] V. Armenio and V. Fiorotto, \The importance of the forces acting on particles in turbulent ows," Phys. Fluids 13, 2437{2440 (2001). [32] M. A. T. van Hinsberg, H. J. H. Clercx, and F. Toschi, \Enhanced settling of nonheavy inertial particles in homogeneous isotropic turbulence: The role of the pressure gradient and the basset history force," Phys. Rev. E 95, 023106 (2017). [33] M. van Aartrijk and H. J. H. Clercx, \Vertical dispersion of light inertial particles in stably strati ed turbulence: The in uence of the basset force," Phys. Fluids 22, 013301 (2010). [34] P. J. Ireland, T. Vaithianathan, P. S. Sukheswalla, B. Ray, and L. R. Collins, \Highly parallel particle- laden ow solver for turbulence research," Comput. Fluids 76, 170 { 177 (2013). [35] M. A. T. van Hinsberg, J. H. M. T. Boonkkamp, F. Toschi, and H. J. H. Clercx, \On the eciency and accuracy of interpolation methods for spectral codes," SIAM J. Sci. Comput. 34, B479{B498 (2012). [36] E.-W. Saw, G. P. Bewley, E. Bodenschatz, S. S. Ray, and J. Bec, \Extreme uctuations of the relative velocities between droplets in turbulent air ow," Phys. Fluids 26, 111702 (2014). [37] M. James and S. S. Ray, \Enhanced droplet collision rates and impact velocities in turbulent ows: The e ect of poly-dispersity and transient phases," Sci. Rep. 7, 12231 (2017). [38] A. Bhatnagar, K. Gustavsson, and D. Mitra, \Statistics of the relative velocity of particles in turbulent ows: Monodisperse particles," Phys. Rev. E 97, 023105 (2018). [39] A. Bhatnagar, K. Gustavsson, B. Mehlig, and D. Mitra, \Relative velocities in bidisperse turbulent aerosols: Simulations and theory," Phys. Rev. E 98, 063107 (2018). [40] P. G. Sa man and J. S. Turner, \On the collision of drops in turbulent clouds," J. Fluid Mech. 1, 1630 (1956). [41] G. Falkovich, A. Fouxon, and M. G. Stepanov, \Acceleration of rain initiation by cloud turbulence," Nature 419, 151{154 (2002). [42] M. Wilkinson, B. Mehlig, and V. Bezuglyy, \Caustic activation of rain showers," Phys. Rev. Lett. 97, 048501 (2006). [43] S. Ravichandran and R. Govindarajan, \Caustics and clustering in the vicinity of a vortex," Phys. Fluids 27, 033305 (2015). [44] J. Bec, A. Celani, M. Cencini, and S. Musacchio, \Clustering and collisions of heavy particles in random smooth ows," Phys. Fluids 17, 073301 (2005). [45] M. Wilkinson, B. Mehlig, S. Ostlund, and K. P. Duncan, \Unmixing in random ows," Phys Fluids 19, 113303 (2007). [46] J. Bec, S. Musacchio, and S. S. Ray, \Sticky elastic collisions," Phys. Rev. E 87, 063013 (2013). [47] K. Gustavsson and B. Mehlig, \Distribution of velocity gradients and rate of caustic formation in turbulent aerosols at nite kubo numbers," Phys. Rev. E 87, 023016 (2013). [48] K. Gustavsson and B. Mehlig, \Relative velocities of inertial particles in turbulent aerosols," J. Turbul. 15, 34{69 (2014). [49] V. E. Perrin and H. J. J. Jonker, \Preferred location of droplet collisions in turbulent ows," Phys. Rev. E 89, 033005 (2014). [50] V. E. Perrin and H. J. J. Jonker, \E ect of the eigenvalues of the velocity gradient tensor on particle collisions," J. Fluid Mech. 792, 3649 (2016). [51] M. Wan, Z. Xiao, C. Meneveau, G. L. Eyink, and S. Chen, \Dissipation-energy ux correlations as evidence for the lagrangian energy cascade in turbulence," Phys. Fluids 22, 061702 (2010). [52] L. Agasthya, J. R. Picardo, S. Ravichandran, R. Govindarajan, and S. S. Ray, \Droplet collisions in turbulence: Insights from a burgers vortex," arXiv e-prints (2018), arXiv:1811.04486. [53] R. Onishi and J. C. Vassilicos, \Collision statistics of inertial particles in two-dimensional homogeneous 10 isotropic turbulence with an inverse cascade," J. Fluid Mech. 745, 279299 (2014). [54] S. Ravichandran and R. Govindarajan, \Vortex-dipole collapse induced by droplet inertia and phase change," J. Fluid Mech. 832, 745776 (2017). [55] A. Liberzon, M. Guala, W. Kinzelbach, and A. Tsinober, \On turbulent kinetic energy production and dissipation in dilute polymer solutions," Phys. Fluids 18, 125101 (2006). [56] P. Perlekar, D. Mitra, and R. Pandit, \Direct numerical simulations of statistically steady, homoge- neous, isotropic uid turbulence with polymer additives," Phys. Rev. E 82, 066313 (2010). [57] J. Bec, H. Homann, and S. S. Ray, \Gravity-driven enhancement of heavy particle clustering in turbulent ow," Phys. Rev. Lett. 112, 184501 (2014).

Journal

Nonlinear Sciences – arXiv (Cornell University)

Published: Oct 24, 2018

There are no references for this article.