Downloaded from rspa.royalsocietypublishing.

org on June 3, 2012

Coiling of viscous jets

N. M. Ribe Proc. R. Soc. Lond. A 2004 460, 3223-3239 doi: 10.1098/rspa.2004.1353

Email alerting service

Receive free email alerts when new articles cite this article - sign up in the box at the top right-hand corner of the article or click here

To subscribe to Proc. R. Soc. Lond. A go to: http://rspa.royalsocietypublishing.org/subscriptions

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

10.1098/rspa.2004.1353

Coiling of viscous jets

By N e i l M. R i b e CNRS UMR 7579, Institut de Physique du Globe, 4 place Jussieu, 75252 Paris CEDEX 05, France (ribe@ipgp.jussieu.fr)
Received 16 January 2004; accepted 10 May 2004; published online 27 July 2004

A stream of viscous uid falling from a sucient height onto a surface forms a series of regular coils. I use a numerical model for a deformable uid thread to predict the coiling frequency as a function of the threads radius, the ow rate, the fall height, and the uid viscosity. Three distinct modes of coiling can occur: viscous (e.g. toothpaste), gravitational (honey falling from a moderate height) and inertial (honey falling from a great height). When inertia is signicant, three states of steady coiling with dierent frequencies can exist over a range of fall heights. The numerically predicted coiling frequencies agree well with experimental measurements in the inertial coiling regime.
Keywords: viscous jet; buckling instability; uid rope coiling

1. Introduction
The periodic buckling of a uid jet incident on a surface is a striking uid mechanical instability with applications from food processing (Tome & McKee 1999) to polymer processing (Pearson 1985) and geophysics (Griths & Turner 1988). Its most beautiful manifestation is the uid rope-coil eect that occurs when a thin stream of honey is poured onto toast (gure 1a). Fluid coiling has been studied extensively in the laboratory for nearly 50 years (Barnes & Woodcock 1958; Barnes & MacKenzie 1959; Cruickshank 1980; Cruickshank & Munson 1981; Huppert 1986; Griths & Turner 1988; Mahadevan et al . 1998). However, its mechanism remains incompletely understood. The rst important theoretical advance was Taylors recognition that uid buckling requires a longitudinal compressive stress, like the buckling of an elastic column under a load (Taylor 1968). Subsequently, the critical fall height and frequency at the onset of coiling were determined using linear stability analysis (Cruickshank 1988; Tchavdarov et al . 1993). Most recently, Mahadevan et al . (1998, 2000) proposed a scaling law for inertia-dominated coiling that agreed well with experimental measurements in the high-frequency limit. In summary, current theoretical understanding of uid coiling is limited to the extremes of very low (incipient coiling) and very high (inertial coiling) frequencies. Here I use a numerical approach to model coiling over the whole frequency range, and to document the existence of three distinct coiling modes with dierent dynamics. The conguration studied is the one used in most laboratory experiments, where uid with density and kinematic viscosity is injected at a volumetric rate Q through a hole of radius a0 and falls a distance H onto a plate (gure 1a). For simplicity I shall neglect surface tension, which has only a small (at most a few per cent) eect on
Proc. R. Soc. Lond. A (2004) 460, 32233239 c 2004 The Royal Society

3223

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3224
Q (a) 2a0

N. M. Ribe

(b)

(c)

(d )

2a1

tail coil

2R
Figure 1. Modes of uid coiling. (a) Coiling of a jet of viscous corn syrup (photograph by the author), showing the parameters of a typical laboratory experiment. (b)(d) Jet shapes calculated using Auto97 (Doedel et al . 2002) for three modes of uid coiling. (b) Viscous coiling (H/a0 = 20, B gH 2 /U0 = 0, Re U0 H/ = 0). (c) Gravitational coiling (H/a0 = 20, B = 100, Re = 0). (d) Inertial coiling (H/a0 = 37, B = 164, Re = 0.031).

the coiling frequency. The jets point of contact with the plate rotates with angular velocity and describes a circle of radius R. In most cases, the jet consists of a long, nearly vertical tail which feeds uid to a coil next to the plate. This study is based on equations that describe the dynamics of a thin viscous jet whose slenderness a 1, where a and are characteristic values of the jets radius and axial curvature, respectively. The equations governing the steady coiling of such a jet are derived in 2, where it is shown that the phenomenon is described by a 17th-order nonlinear two-point boundary-value problem with two free parameters and 19 boundary conditions. The numerical solution of these equations is explained in 3. Readers uninterested in the details of the derivation may skip directly to the presentation of the results in 4.

2. Governing equations for steady coiling

The essential goal of the derivation that follows is to exploit the jets slenderness to reduce the three-dimensional NavierStokes equations to one-dimensional equations that describe the dynamics of a curved line (the jet axis) endowed with nite resisProc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

d3(s) d2(s)

3225

x(s) a(s)

d1(s)

e3 s e2 e1
Figure 2. Geometry of a viscous jet. The Cartesian coordinates of the jets axis relative to an arbitrary origin, O, are x(s), where s is the arc length along the axis. The jets radius is a(s). The unit tangent vector to the jet axis is d3 (s) x , and d1 (s) and d2 (s) d3 d1 are material unit vectors in the plane of the jets cross-section.

tance to stretching, bending and twisting. The problem is further reduced to one of steady motion by working in a reference frame that rotates with the coil. All the dependent variables are then functions only of the arc length s along the jet axis, which ranges from s = 0 (the injection point) to s = (the unknown point of contact with the plate). In the following, dierentiation with respect to s will be denoted by a prime. Unless otherwise noted, all Latin indices range over the values 1, 2 and 3, while Greek indices range over 1 and 2 only. The Einstein summation convention over repeated indices (both Latin and Greek) is assumed throughout. Figure 2 shows an element of a slender viscous jet with variable radius a(s). The Cartesian coordinates of the jets axis relative to unit vectors ei rotating with the coil are x(s), dened such that (x1 , x2 , x3 ) = (0, 0, 0) is the point where the uid is injected. The vector e3 points up, opposite to the gravitational acceleration f g e3 . At each point on the jets axis, a triad of orthogonal unit vectors is dened, comprising the tangent vector d3 (s) x and two vectors d1 (s) and d2 (s) d3 d1 in the plane of the jets cross-section. The vectors d1 and d2 are arbitrary, but are assumed to be material vectors that rotate with the uid. An alternative approach in which d1 and d2 are dened geometrically as the principal normal and binormal of the jet axis (Entov & Yarin 1984) leads in my experience to equations that are numerically unstable when the total curvature of the axis is small. Let y1 and y2 be orthogonal coordinates normal to the jet axis in the directions d1 and d2 , respectively. The Cartesian coordinates of an arbitrary point within the jet are then r (s, y1 , y2 ) = x(s) + y1 d1 (s) + y2 d2 (s) x + y .
Proc. R. Soc. Lond. A (2004)

(2.1)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3226

N. M. Ribe

The orientation of the local basis vectors relative to the Cartesian basis is described by the matrix of direction cosines 2 2 2 2 q1 q2 q3 + q0 2(q1 q2 + q0 q3 ) 2(q1 q3 q0 q2 ) 2 2 2 2 q1 + q2 q3 + q0 2(q2 q3 + q0 q1 ) , (2.2) dij di ej = 2(q1 q2 q0 q3 ) 2 2 2 2 2(q2 q3 q0 q1 ) q1 q2 + q3 + q0 2(q1 q3 + q0 q2 ) where qi (s) are Euler parameters (Whittaker 1944) satisfying
2 2 2 2 + q1 + q2 + q3 = 1. q0

(2.3)

Use of the Euler parameters avoids the polar singularities associated with the more familiar Eulerian angles. The ordinary dierential equations satised by xi and qj are (Mahadevan & Keller 1996) (2.4) xi = d3i , 1 q0 = 2 (1 q1 2 q2 3 q3 ), ( q q + q ) , q1 = 1 1 0 2 3 3 2 2 (2.5) q2 = 1 2 (1 q3 + 2 q0 3 q1 ), ( q + q + q ) , q3 = 1 1 2 2 1 3 0 2 where i di is the curvature vector that measures the rates of change of the local basis vectors along the jet axis according to the generalized Frenet relations di = di . (2.6) To rst order in the lateral coordinates y1 and y2 , the velocity of a uid particle in the jet relative to the rotating reference frame is u = U d3 1 2 U y + y, where (2.8) = 1 U d1 + 2 U d2 + 3 d3 is one-half the vorticity at the jet axis y1 = y2 = 0, U (s)d3 (s) is the velocity along the axis, and 3 (s) is the angular velocity (spin) of the uid about the axis. The second term on the right-hand side of (2.7) is the lateral velocity induced by stretching of the axis at a rate U . The third term represents the velocity associated with bending and twisting of the jet. Because the base vectors di are convected with the uid, their angular velocity as they travel along the jet axis is the sum of the angular velocity of the ow and any additional spin that is imparted to the vectors d1 and d2 when they are injected at s = 0. Now d1 and d2 can only be steady in the rotating frame if they are injected at s = 0 in such a way as to follow the rotation of the jet as a whole. This is equivalent to imparting to d1 and d2 and additional spin of magnitude , where the minus sign accounts for the fact that e3 points up and d3 (0) down. The evolution equation for di is therefore (2.9) U di = ( d3 ) di , where the left-hand side is the (steady) convective rate of change of di along the jet axis. Substitution of the Frenet relations (2.6) into (2.9) yields the fundamental condition for the steadiness of di (s): 3 = U 1 (3 ). Equation (2.10) allows 3 to be eliminated from all the equations that follow.
Proc. R. Soc. Lond. A (2004)

(2.7)

(2.10)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

3227

Equations for the global balance of force and torque on the jet are obtained by integrating the NavierStokes equations over the jets cross-section S (Appendix A). The dynamical variables that then appear are the stress resultant vector N Ni di =
S

dS

(2.11)

and the bending/twisting moment vector M Mi di =

S

y dS,

(2.12)

where is the stress vector acting on the jets cross-section. The quantities N3 , M1 , M2 and M3 measure the jets resistance to stretching, bending in two orthogonal directions, and twisting, respectively. The resultants N1 and N2 are the integrals of the shear stresses that accompany bending and twisting, and are generally small. The integrated balance of forces per unit jet length is (Appendix A) A[ ( x) + 2 U + U U ] = N + Af , (2.13) where A(s) a(s)2 is the area of the jets cross-section. The two terms on the right-hand side represent the viscous force that resists deformation of the jet and the force of gravity, respectively. The three inertial terms on the left-hand side represent the centrifugal force, the Coriolis force and the accelerations due to variations in the axial velocity U U d3 , respectively. The integrated torque balance is I K = M + d3 N + I [(f d3 ) ( f )d3 ],
4

(2.14)

where and I a /4 is the moment of inertia of the jets cross-section and the components of K Ki di are K = U (U ) U d3 2 d3 x +
3 [d 3 (d33

+ 23 ) + U ( + 3 )],
2 3 U d3 .

(2.15 a) (2.15 b)

K3 = 2U 3 2U (d33 + 3 ) + d x + 4

Four additional dierential equations appear in the form of constitutive relations for the stress resultant N3 and the moments M1 , M2 , and M3 . The derivation outlined in Appendix B yields N3 = 3AU , M1 = 3I [(U 1 ) + 2 (3 3 U )], M2 = 3I [(U 2 ) 1 (3 3 U )], M3 = 2I3 , (2.16 a) (2.16 b) (2.16 c) (2.16 d)

where is the dynamic viscosity. Finally, the system of equations is closed by eliminating the jet radius a using the volume ux conservation relation a2 U = Q. (2.17)

Equations (2.4), (2.5), (2.13), (2.14) and (2.16) are a system of 17 rst-order dierential equations for the 17 variables x1 , x2 , x3 , q0 , q1 , q2 , q3 , 1 , 2 , U , 3 ,
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3228

N. M. Ribe

N1 , N2 , N3 , M1 , M2 and M3 . However, there are also two unknown parameters ( and ), so 19 boundary conditions are required. Dening the point of injection as the origin yields the three conditions x1 (0) = x2 (0) = x3 (0) = 0. The jet is injected vertically downward, which requires d33 (0) = 1 or q0 (0) = q3 (0) = 0. (2.19) The vanishing of the rotation rate of the tangent vector d3 at the point of injection requires (2.20) 1 (0) = 2 (0) = 0. The velocity of injection is U0 , or U (0) = U0 . (2.21) If the jet is extruded without rotation in the laboratory frame, its apparent rate of rotation viewed in the rotating frame (noting that e3 points up and d3 (0) down) is 3 (0) = . (2.22) Consider now the contact point s = , which may be supposed, without loss of generality, to lie on the positive x1 axis in the rotating frame, so that x2 ( ) = 0. Contact of the jet with the plate requires x3 ( ) = H + a( ). (2.24) The values of the Euler parameters at s = can be found by noting that the contact point can move only if both the tangent and the principal normal to the jet axis are horizontal there, or d33 ( ) = 2 ( )d13 ( ) 1 ( )d23 ( ) = 0. (2.25) Without loss of generality we may take d1 ( ) to be the principal normal, which points in the e1 direction (toward the centre of the coil). Moreover, the tangent vector d3 ( ) points in the e2 direction, implying d32 ( ) = d23 ( ) = d11 ( ) = 1. Accordingly, the moving contact line conditions (2.25) are satised if q0 ( ) = q1 ( ) = q2 ( ) 21/2 = q3 ( ) + 21/2 = 0 and 1 ( ) = 0. The contact point traces out a circle with radius 1/2 ( ), implying x1 ( )2 ( ) = 1. (2.28) The no-slip condition at the contact point implies that the jet cannot rotate about its axis, or (2.29) 3 ( ) = 0. Finally, the axial uid velocity at the contact point must equal the velocity of the contact itself, or (2.30) 2 ( )U ( ) = . Equations (2.18)(2.24) and (2.26)(2.30) are the 19 boundary conditions required.
Proc. R. Soc. Lond. A (2004)

(2.18)

(2.23)

(2.26) (2.27)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

3229

3. Numerical solution
I solve the above two-point boundary-value problem numerically using the program Auto97 (Doedel et al . 2002). This program implements an automatic continuation (homotopy) method, wherein a simple analytical solution of the governing equations is gradually adjusted until it satises all the boundary conditions. The starting point is the following analytical solution of the equations for a (non-coiling) jet having the form of a quarter circle in the absence of gravity and inertia: = 0, =1 2 H, s s , x3 = H sin , x1 = H 1 cos 2 2 (3.1) s s 1 2 = H , U = U0 , q3 = sin , q1 = cos , 4 4 x2 = q0 = q2 = 1 = 3 = N1 = N2 = N3 = M1 = M2 = M3 = 0. The above solution satises all but ve of the boundary conditions. Accordingly, I rewrite these conditions by adding continuation parameters ci (i = 1, . . . , 5), as follows: 0 = H2 (0) 1 + c1 = H2 ( )U ( ) U0 + c2 (U0 H ) = x3 ( ) + H c3 a( ) = q0 ( ) 21/2 cos + = q1 ( ) 21/2 sin = q2 ( ) 21/2 cos = q3 ( ) 21/2 sin + , where = 1 4 (c4 c5 2). (3.2) (3.3)

The analytical solution (3.1) satises the boundary conditions (3.2) with ci = 0, while the boundary conditions for the full coiling problem are obtained when ci = 1. Beginning from the analytical solution with ci = 0, the numerical procedure consists in gradually increasing the ci until a solution of the full coiling problem is reached. This solution is then continued further by adding gravity and inertia terms. Before numerical solution, the equations and boundary conditions are non-dimensionalized using the scales H (for xi ), H 1 (for 1 and 2 ), U0 (for U ), U0 /H (for 4 2 (for s). The dimensionless arc 3 ), a2 0 U0 /H (for Ni ), a0 U0 /H (for Mi ) and length s = s/ [0, 1], as required by Auto97. The resulting system of equations involves the two unknown parameters /H and H/U0 and the three dimensionless groups a0 /H (slenderness), gH 2 /U0 B (gravity number) and U0 H/ Re (Reynolds number).

4. Scaling regimes
The motion of a coiling jet is controlled by the balance among viscous forces, gravity and inertia. Viscous forces arise from internal deformation of the jet by stretching (localized mainly in the tail) and by bending and twisting (mainly in the coil). Inertia includes the usual centrifugal and Coriolis accelerations, as well as terms proportional
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3230

N. M. Ribe

to the along-axis rate of change of the magnitude and direction of the axial velocity U d3 . The coiling frequency is determined by the balance of forces in the coil itself, where the jet radius a a1 and the associated axial speed U1 Q/a2 1 are nearly constant (gure 1a). To estimate the magnitudes of the forces in the coil, I use the torque-balance equation (2.14) to eliminate the shear stress resultants N from the force-balance equation (2.13). This yields 0=
3 M

+ A

3 U

gAd3 + ,

(4.1)

where the ellipsis indicates additional terms that are of the same or higher order in the slenderness as those shown. The three terms in (4.1) represent the viscous force, inertia, and the gravitational force, respectively, all per unit length of the jet axis. Noting that d/ds R1 in the coil and using the constitutive relations (2.16 b) and (2.16 c) for M , one nds
2 1 inertia a2 . 1 U1 R (4.2) Three dierent modes of coiling are possible, depending on how the viscous forces in the coil are balanced. The simplest case (viscous coiling) occurs when gravity and inertia are both negligible and the net viscous force acting on any element of uid is zero. Coiling is here driven entirely by the injection of the uid, like toothpaste squeezed from a tube. Because the jet deforms by bending and twisting with negligible stretching, its radius is nearly constant (gure 1b). Therefore, a1 a0 and U1 U0 . Moreover, the uid velocity is entirely controlled by the injection speed, and is independent of the jets viscosity and radius. Dimensional considerations and the general relation U1 /R then imply 4 , viscous force a4 1 U1 R

gravitational force ga2 1,

R H RV ,

2 1 a V . 1 QH

(4.3 a)

A second mode, gravitational coiling, occurs when viscous forces are balanced by gravity. The jet now comprises a long tapering tail and a coil that occupies only a small portion of the total height H (gure 1c). The scaling laws for this mode are R g 1/4 1/4 Q1/4 RG ,
2 3 /4 g 1/4 1/4 a G . 1 Q

(4.3 b)

A third mode, inertial coiling, occurs when viscous forces in the coil are balanced by inertia (gure 1d). The scaling laws for inertial coiling are R 1/3 a1 Q1/3 RI ,
4 /3

1/3 a1

10/3

Q4/3 I ,

(4.3 c)

and were rst proposed by Mahadevan et al . (2000). It may at rst sight seem contra1, dictory that inertia can be important when the Reynolds number Re = U0 H/ as is the case for the jet shown in gure 1d (Re = 0.031). The paradox is resolved by noting that the eective Reynolds numbers in the coil and in the tail of the jet may be very dierent. Indeed, Re is an appropriate measure of the ratio of inertia to viscous forces only in the tail, and moreover only when U1 does not greatly exceed 2 1 2 U0 ; then, inertia a2 and the viscous force a2 . In the coil itself, 0 U0 H 0 U0 H the magnitudes of inertia and the viscous force are given by (4.2), and the ratio of the two is of the order of unity in the inertial coiling regime dened by the scaling laws (4.3 c). The physical reason for the larger role of inertia in the coil is that, for a
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

10 1 5 1

3231

/ V
2 1 0.5 1 2

G / V

10

20

Figure 3. Scaling law for slow (inertia-free) coiling. Frequencies predicted by Auto97 are shown for H/a0 = 20 (solid line), 40 (long-dashed line) and 80 (short-dashed line). V and G are dened by equations (4.3 a) and (4.3 b), respectively.

given strain rate, the viscous forces within a thin lament deformed by bending are much smaller than in one deformed by stretching. I now demonstrate the existence of the three coiling modes by solving the full 17th-order boundary-value problem, using the numerical method described in 3. Consider rst the case of slow coiling, which includes the two modes (viscous and gravitational) that involve no inertia. As the importance of gravity increases relative to the viscous forcesas the height of fall increases, for examplea transition from viscous to gravitational coiling will occur. The control parameter for this transition will be the ratio of the characteristic frequencies of the two modes, or G /V H (g/Q)1/4 . Accordingly, a loglog plot of /V versus G /V should dene a universal curve with two distinct legs: one with zero slope corresponding to viscous coiling and another with unit slope corresponding to gravitational coiling. To test this, I used Auto97 to determine the coiling frequency as a function of the gravity number gH 2 /U0 B for three values of H/a0 , with all inertial terms suppressed (Re U0 H/ = 0). For each solution, the jet radius a1 was calculated as the average radius of those portions of the jet where the rate of energy dissipation due to stretching was less than 5% of the total. The scaled frequencies (gure 3) do in fact display the expected two-leg structure, with a transition from viscous to gravitational coiling occurring in the interval 2 < G /V < 3. The three curves shown in gure 3 dier slightly because the jet axis at the contact point s = is at a small but nite height a( ) a0 above the plate (see gure 1b). A truly universal curve is obtained in the limit H/a0 , and is indistinguishable from the curve for H/a0 = 80 in gure 3. The eects of inertia are complex, and are best introduced via a direct comparison of numerical predictions with laboratory experiments. Mahadevan et al . (1998) measured the coiling frequency and the jet radius a1 for viscous silicone oil ( = 1000 cm2 s1 ) by varying the fall height H with Q and a0 xed. Figure 4 shows curves of frequency versus height predicted numerically using Auto97 for the three pairs of values of Q and a0 used by Mahadevan et al . (1998), together with the 15 frequencies they measured (open symbols). The experimental values of H , which
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3232
500

N. M. Ribe

200

100

(s1)

50 4 20 3

10 1 5 5 10

20 H (cm)

50

Figure 4. Coiling frequency versus fall height in the presence of inertia. Experimental measurements (open symbols) are shown for three series of experiments by Mahadevan et al . (1998) with viscous silicone oil ( = 1000 cm2 s1 ) and {Q, a0 } = {0.57 cm3 s1 , 0.32 cm} (circles), {0.98 cm3 s1 , 0.40 cm} (squares) and {1.31 cm3 s1 , 0.48 cm} (triangles). The solid, dashed and dotted lines are the frequencyheight curves predicted by Auto97 for the same three pairs of values of Q and a0 . Labels 15 denote reference points along the curves.

Mahadevan et al . (1998) did not report, were obtained by comparing the values of a1 predicted by Auto97 with those measured by Mahadevan et al . (1998). The numerical predictions agree with the measured frequencies to within a root-mean-squared (RMS) error of 26%. The most striking feature of the frequencyheight curves in gure 4 is their multivaluedness in the height range 1115 cm, where three coiling states with dierent frequencies are possible for the same values of a0 , H and Q. This complex structure reects a transition from gravitational to inertial coiling as the fall height increases. Note rst from (4.3 b) and (4.3 c) that the frequencies of gravitational and inertial coiling do not depend explicitly on the fall height H , but only on Q, a1 , and g . Standard dimensional analysis then implies that I =F , , G G = , g 1/5 Q3/5 (4.4)

where F is an unknown function to be determined. The rst argument of F is the control parameter for the transition from gravitational to inertial coiling, while the second remains constant during an experiment in which the fall height (and hence a1 ) varies.
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

3233

1 1.0 1 5

/ G

0.5

3 4

2 0.2 0.5 1 2 I / G 5

Figure 5. Transition from gravitational to inertial coiling in numerical solutions corresponding to the experiments of Mahadevan et al . (1998). Solid, dashed and dotted curves are rescaled versions of the same curves in gure 3. Values of /g 1/5 Q3/5 for these three curves are 366, 265 and 223, respectively. The leftmost (zero slope) and rightmost (unit slope) limits of the curves correspond to purely gravitational and purely inertial coiling, respectively. Reference labels 15 are the same as in gure 4.

Figure 5 shows a loglog plot of /G versus I /G for the numerical simulations of gure 4. The three curves coincide exactly at both the left and right limits of the 1), the curves approach plot, indicating scaling universality. On the left (I /G a limit = 0.49G that corresponds to purely gravitational coiling, and which is identical to the rightmost portion of the curve in gure 3. At the far right of gure 5 1), the limiting form is = 0.185I , corresponding to purely inertial (I /G coiling. In between is a complex transitional region in which both gravity and inertia are signicant. In this region, /G depends not only on I /G but also on , which ranges from 223 to 366 for the three curves shown. Between reference points 1 and 2, the coiling is dominantly gravitational, but its frequency is substantially reduced (up to a factor of 2.3) by inertia. Between points 2 and 3, the frequency is a strongly decreasing function of height (cf. gure 4), reecting the increase of a1 with height over this interval. At the end of the transitional region (point 4), the coiling is dominantly inertial, but its frequency is increased (by up to 25%) by gravity. The eect of gravity progressively diminishes until pure inertial coiling (unit slope in 1. gure 5) is attained in the limit I /G

5. Dynamics of the tail

To this point, the jet radius a1 in the coil has been treated as an independent variable. In reality, a1 is controlled by the amount of gravity-induced stretching that occurs in the tail, and therefore depends on the external parameters (H , Q, a0 , and g ) of the experiment. This dependence can be determined using a simple model for steady
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3234

N. M. Ribe

1.00 0.50

100 4 1 10

a1/a0

0.20 0.10 1 0.05

0.02

101

100

101 102 2 / U gH1 0

103

104

Figure 6. Total thinning a1 /a0 due to unidirectional stretching in the jet tail of length H1 . Portions of the curves with slope 1/4 correspond to free fall with negligible viscous resistance.

unidirectional stretching. The governing equation for this is just the d3 -component of the global force-balance equation (2.13) with 1 = 2 = x1 = x2 = 0, or 3U (U 1 U ) + g U U = 0. (5.1)

The three terms in (5.1) represent the viscous resistance to stretching, gravity and inertia, respectively. The boundary conditions are U (0) U0 = U (H1 ) = 0, (5.2)

where the eective length H1 < H of the tail is dened by the point at which the stretching rate U vanishes. I solved (5.1) and (5.2) numerically using a relaxation method (Press et al . 1996). Figure 6 shows a1 /a0 (U0 /U1 )1/2 as a function of the 2 /U0 and the Reynolds number Re1 = U0 H1 / . eective gravity number B1 = gH1 Gravity and inertia have opposite eects on thinning, enhancing and inhibiting it, respectively. The solutions shown in gure 6 have simple analytical forms in two limiting cases. The rst is Re1 = 0, corresponding to a balance of gravity and viscous forces with negligible inertia. The solution for this case is U = U1 cos
2

2B1 U0 3U1

1 /2

s H1 2H1

(5.3)

where U1 satises the transcendental equation U (0) = U0 that corresponds to the boundary condition at the injection point. In the limit B1 , the asymptotic expression for a1 /a0 (U0 /U1 )1/2 is 1/2 a1 6 1/2 3/2 1 = [1 6B1 + 6B1 + O(B1 )]. (5.4) B1 a0 2
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

3235

3 / 2 . Inertia now balances gravity everywhere The second limit is Re1 B1 gH1 except in thin boundary layers near the injection point and the lower end of the tail, and the solution is 1/4 1/4 a1 2gH1 2B1 = . (5.5) 2 a0 U0 Re1

This solution describes thinning by free fall with negligible injected kinetic energy, and corresponds to the portions of the curves with slope 1/4 in gure 6. To compare the values of a1 /a0 for the simple stretching model (gure 6) with those predicted by solutions of the complete 17th-order boundary-value problem, one needs to estimate the eective length H1 of the tail in the latter solutions. I assume H1 = H R as a simple approximation, but the results are hardly sensitive to the exact choice. The predictions of gure 6 then agree closely (RMS error 5%) with all the values of a1 /a0 predicted by Auto97 that were used to construct gures 35.

6. Discussion
Fluid coiling has fundamental similarities with two related phenomena: periodic folding of viscous sheets and coiling of elastic ropes. Folding of viscous sheets is easily demonstrated in the home kitchen using honey, molten chocolate or cake batter, and may occur in the Earths mantle when subducted oceanic lithosphere impinges on a boundary between layers with dierent viscosities and/or densities (Griths & Turner 1988). The two-dimensional ow within a folding sheet is simpler than that within a coiling lament, involving only two modes of deformation (stretching and bending) rather than four. On the other hand, folding, unlike coiling, is an essentially time-dependent phenomenon in which the motion of the contact line is both nonuniform and discontinuous. The folding of viscous sheets and laments was studied by Skorobogatiy & Mahadevan (2000), who showed that experimentally measured folding frequencies of laments conned to a plane obeyed a scaling law having the same form as the gravitational coiling law (4.3 b). More recently, Ribe (2003) used a numerical model to show that a folding viscous sheet exhibits a transition from viscous to gravitational folding very similar to the one documented here (gure 3) for coiling. Coiling of an elastic rope has been studied by Mahadevan & Keller (1996) (henceforth MK96), using a numerical model for a thin rope moving under the action of gravity, inertia and the elastic forces that resist bending. Because deformation by stretching and twisting is negligible in the rope, its coiling is described by a two-point boundary problem of order 13, compared with order 17 for uid coiling. MK96 show that four modes of coiling are possible, depending on the values of the dimensionless parameters AgH 3 U2 , = , (6.1) F = gH EI where U is the (constant) feeding velocity and E is the Youngs modulus of the rope material. When F = = 0, both gravity and inertia are negligible, and the net elastic force on any element of the rope is zero (gure 1 of MK96). This case is analogous to the viscous mode of uid coiling. For F 1 and F 1, gravity-dominated coiling occurs, in which the rope is nearly vertical over most of its length (the tail) and gravity balances elastic forces in the coil (gure 4b and leftmost portions of gures 2
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3236

N. M. Ribe

and 3 of MK96). This case is therefore analogous to the gravitational mode of uid coiling. The two remaining modes, which MK96 call inertia-dominated and inertia- and gravity-dominated, have no analogues in uid coiling. In both cases, elastic forces are important only in small boundary layers near the feeding point and the plate, and most of the coil behaves as a perfectly exible string in which axial tension is balanced by gravity and inertia. On the other hand, an elastic rope has no mode analogous to the inertial coiling of a uid jet, in which inertia balances viscous resistance to bending in the coil. A nal important dierence between elastic and uid coiling is that the frequency of the former is always single valued, whereas three states of uid coiling with different frequencies are possible for the same input parameters under certain conditions (cf. gures 3 and 4). But it is not guaranteed that all three of these solutions are stable. Theoretical analysis of this question would require a more general timedependent theory for viscous jets, and is beyond the scope of this paper. However, recent experiments (Maleki et al . 2004) suggest that the solutions with the lowest and highest frequencies are stable, while the one with intermediate frequency (i.e. between reference points 2 and 3 in gures 3 and 4) is unstable. While there is good agreement between the numerically predicted coiling frequencies and those measured by Mahadevan et al . (1998), further experimental conrmation of the numerical model is clearly needed. The most extensive experimental study of uid coiling is that of Cruickshank (1980), who presented in graphical form some 700 measurements of the coiling frequency for dierent values of , a0 , Q and H . Regrettably, Cruickshank (1980) did not measure the critical parameter a1 , an oversight which reduces greatly the utility of his data. New experiments that demonstrate the existence of the three coiling modes identied here have now been performed, and will be reported separately (Maleki et al . 2004).
This research was supported by the Centre National de la Recherche Scientique (France). I thank A. Boudaoud, D. Bonn, M. Brenner, J. Bush, C. Clanet, A. Davaille, R. Golestanian, H. Huppert, J. Lister, M. Maleki, H. Stone, M. G. Worster and W. Zhang for helpful discussions. J. Hinch, H. Huppert, S. Morris and an anonymous referee suggested many improvements to the original presentation. L. Mahadevan kindly made available the experimental data from Mahadevan et al . (1998). The sugar syrup used for the experiment shown in gure 1a was generously supplied by SYRAL.

Appendix A. Global force and torque balance

A rigorous derivation of the governing equations for the ow within the jet requires the use of general nonorthogonal coordinates dened by the variable transformation (2.1). The general approach I follow here is that of Green & Zerna (1992), henceforth abbreviated as GZ. In this appendix, i denotes partial dierentiation with respect to the variable yi , where y3 s. Direct partial dierentiation of (2.1) yields the covariant basis vectors g1 1 r = d1 , where g2 2 r = d2 , g3 3 r = hd3 3 (y2 d1 y1 d2 ), (A 1) (A 2)
i i j ,

h g1 (g2 g3 ) = 1 2 y1 + 1 y2 .
i

The associated contravariant (reciprocal) base vectors g satisfy g gj = g = d1 + h

1 1

whence (A 3)

3 y2 d3 ,

g = d2 h
2

3 y1 d3 ,

g =h

d3 .

Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets The strain rate tensor within the jet is (GZ, p. 148) eij = 1 2 (gi j u + gj i u). Incompressibility of the uid requires g ij eij = 0 with g ij = g i g j , or 1 (hu1 ) + 2 (hu2 ) + 3 u3 + 3 (y2 1 u3 y1 2 u3 ) = 0. The stress tensor relative to the local basis vectors gi is ij = pg ij + 2g ik g il ekl , where p is the pressure. The equations of equilibrium are (GZ, p. 150) = i ( ij dj ) + hf , hr is the acceleration and where r ij = ij = h ik gk dj

3237

(A 4)

(A 5)

(A 6)

(A 7)

(A 8)

is a modied (and non-symmetric) stress tensor relative to the axial basis vectors di di and to a unit area of a reference surface at y1 = y2 = 0. Unlike ij , ij can meaningfully be integrated over cross-sections, because the basis vectors and the surface to which it is referred do not vary across the jet. The equation for global force balance is obtained by integrating the momentum equations (A 7) over cross-sections of the jet and using the denitions (2.11) and (2.12) with = 3i di , yielding
S

= N + Af . hr

(A 9)

The equation for global torque balance is obtained by applying the operator y to (A 7) and then integrating, yielding
S

= M + d3 N + I [(f d3 ) ( f )d3 ]. hy r

(A 10)

The last term in (A 10), which represents a gravity-induced torque, is non-zero for the purely geometrical reason that the centre of gravity of an element of the jet does not lie on the axis if the jet is curved. It remains only to evaluate the integrals on the left-hand sides of (A 9) and (A 10). To rst order in y1 and y2 , the acceleration associated with the velocity eld (2.7) is = [ (x + y )] + 2 u r + U U + U y + ( y )3 U 3 U d3 y
2 2 1 +1 + [(y ) d3 ](U U U U ) [3 2 U U 4 (U ) ]y .

(A 11)

The rst two terms in the above expression are the usual centrifugal and Coriolis forces associated with the rotating reference frame, and the sum of the remaining terms is the acceleration of a particle measured in the rotating frame itself. The nal forms (2.13) and (2.14) of the equations of global force and torque balance are obtained by substituting (A 11) into (A 9) and (A 10) and evaluating the integrals, after eliminating 3 using (2.10).
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

3238

N. M. Ribe

Appendix B. Constitutive relations

The constitutive relations (2.16) can be derived most simply by considering separately deformations dominated by stretching and by bending plus twisting. In all the expressions below, terms of higher order in the slenderness have been neglected. For deformation dominated by stretching, the velocity within the jet is approximately (B 1) u U d3 1 2 U y. The denition (A 8) then implies 33 p + 2U , 11 = 22 p U , (B 2)

where p is constant over the jets cross-section. The vanishing of the normal stress at the jets free surface requires p = U , which implies 33 3U . Substitution of this result into the denition (2.11) yields the constitutive relation (2.16 a). For deformation dominated by bending and twisting, the velocity components u1 and u2 must be expanded to second order in the lateral coordinates y using the continuity equation (A 5). The resulting velocity eld is u U [1 2 y1 + 1 y2 ]d3 + 3 d3 y
2 2 y2 )(B1 d1 B2 d2 ) + 2y1 y2 (B2 d1 + B1 d2 ), + (y1

(B 3) (B 4)

where B = 1 4[
3 (U )

(3 U 3 )],

and terms proportional to the stretching rate U (assumed negligible) have been ignored. The pressure varies linearly across the jet to lowest order, and can be written as p = p (s)y . (B 5) The relevant components of the stress tensor ij are now 3 3 3 y , 33 (p + 8B )y , 11 = 22 (p 4B )y . The vanishing of the normal stress on the jets free surface requires p = 4B . (B 7)

(B 6)

Substitution of the above expressions into the denition (2.12) yields the constitutive relations (2.16 b)(2.16 d). A more rigorous but lengthier derivation based on asymptotic expansions of the three-dimensional NavierStokes equations in powers of the slenderness (e.g. Ribe 2002) gives exactly the same results.

References
Barnes, G. & MacKenzie, J. 1959 Height of fall versus frequency in liquid rope-coil eect. Am. J. Phys. 27, 112115. Barnes, G. & Woodcock, R. 1958 Liquid rope-coil eect. Am. J. Phys. 26, 205209.
Proc. R. Soc. Lond. A (2004)

Downloaded from rspa.royalsocietypublishing.org on June 3, 2012

Coiling of viscous jets

3239

Cruickshank, J. O. 1980 Viscous uid buckling: a theoretical and experimental analysis with extensions to general uid stability. PhD thesis, Iowa State University, Ames, IA, USA. Cruickshank, J. O. 1988 Low-Reynolds-number instabilities in stagnating jet ows. J. Fluid Mech. 193, 111127. Cruickshank, J. O. & Munson, B. R. 1981 Viscous uid buckling of plane and axisymmetric jets. J. Fluid Mech. 113, 221239. Doedel, E., Champneys, A. R., Fairgrieve, T. F., Kuznetsov, Y. A., Sandstede, B. & Wang, X. J. 2002 Auto97: continuation and bifurcation software for ordinary dierential equations. (Available at http://indy.cs.concordia.ca/auto/.) Entov, V. M. & Yarin, A. L. 1984 The dynamics of thin liquid jets in air. J. Fluid Mech. 140, 91111. Green, A. E. & Zerna, W. 1992 Theoretical elasticity. New York: Dover. Griths, R. W. & Turner, J. S. 1988 Folding of viscous plumes impinging on a density or viscosity interface. Geophys. J. 95, 397419. Huppert, H. E. 1986 The intrusion of uid mechanics into geology. J. Fluid Mech. 173, 557594. Mahadevan, L. & Keller, J. B. 1996 Coiling of exible ropes. Proc. R. Soc. Lond. A 452, 1679 1694. Mahadevan, L., Ryu, W. S. & Samuel, A. D. T. 1998 Fluid rope trick investigated. Nature 392, 140. Mahadevan, L., Ryu, W. S. & Samuel, A. D. T. 2000 Correction: uid rope trick investigated. Nature 403, 502. Maleki, M., Habibi, M., Golestanian, R., Ribe, N. & Bonn, D. 2004 Liquid rope coiling. (Submitted.) Pearson, J. R. A. 1985 Mechanics of polymer processing. Elsevier. Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. 1996 Numerical recipes in Fortran77: the art of scientic computing, 2nd edn. Cambridge University Press. Ribe, N. M. 2002 A general theory for the dynamics of thin viscous sheets. J. Fluid Mech. 457, 255283. Ribe, N. M. 2003 Periodic folding of viscous sheets. Phys. Rev. E 68, 036305. Skorobogatiy, M. & Mahadevan, L. 2000 Folding of viscous sheets and laments. Europhys. Lett. 52, 532538. Taylor, G. I. 1968 Instability of jets, threads, and sheets of viscous uid. In Proc. 12th Int. Cong. Applied Mechanics. Springer. Tchavdarov, B., Yarin, A. L. & Radev, S. 1993 Buckling of thin liquid jets. J. Fluid Mech. 253, 593615. Tome, M. F. & McKee, S. 1999 Numerical simulation of viscous ow: buckling of planar jets. Int. J. Numer. Meth. Fluids 29, 705718. Whittaker, E. T. 1944 A treatise on the analytical dynamics of particles and rigid bodies, 4th edn. New York: Dover.

Proc. R. Soc. Lond. A (2004)