SURFACE TENSION MODULE by John W. M.

Bush Department of Mathematics, MIT This set of notes has been developed as supporting material for the Surface Tension module in the 1.63J/2.21J Fluid Dynamics course, which will be presented in a series of 6 lectures at the end of the spring term. These lecture notes have been drawn from many sources, including text books, journal articles, and lecture notes from courses taken by the author as a student. These notes are not intended as a complete discussion of the subject, or as a scholarly work in which all relevant references are cited. Rather, they are intended as an introduction that will hopefully motivate the interested student to learn more about the subject. Topics have been chosen according to their perceived value in developing the physical insight of the students. LECTURE 1: The denition and scaling of surface tension 1.1 Surface tension: a working denition Discussions of the molecular origins of surface or interfacial tension may be found elsewhere (e.g. Israelachvili 1995, Rowlinson & Widom 1982). Our discussion follows that of De Gennes, Brochard-Wyart & Qu er e (2003). Molecules in a uid feel a mutual attraction. When this attractive force is overcome by thermal agitation, the molecules pass into a gaseous phase. Let us rst consider a free surface, for example that between air and water. A water molecule in the uid bulk is surrounded by attractive neighbours, while a molecule at the surface is attracted by a reduced number of neighbours and so in an energetically unfavourable state. The creation of new surface is thus energetically costly, and a uid system will act to minimize surface areas. It is thus that small uid bodies tend to evolve into spheres; for example, a thin uid jet emerging from your kitchen sink will generally pinch o into spherical drops in order to minimize the total surface area (see Lecture 5). If U is the total cohesive energy per molecule, then a molecule at a at surface will lose U/2. Surface tension is a direct measure of this energy loss per unit area of surface. If the characteristic molecular dimension is R and its area thus R2 , then the surface tension is U/(2R2 ). Note that surface tension increases as the intermolecular attraction increases and the molecular size decreases. For most oils, 20 dynes/cm, while for water, 70 dynes/cm. The highest surface tensions are for liquid metals; for example, liquid mercury has 500 dynes/cm.

The origins of interfacial tension are analogous. Interfacial tension is a material property of a uid-uid interface whose origins lie in the dierent attractive intermolecular forces that act in the two uid phases. The result is an interfacial energy per area that acts to resist the creation of new interface, and that is equivalent to a line tension acting in all directions parallel to the interface. Fluids between which no interfacial tension arises are said to be miscible. For example, salt molecules will diuse freely across a boundary between fresh and saltwater; consequently, these uids are miscible, and there is no interfacial tension between them. Our discussion will be conned to immiscible uid-uid interfaces (or uid surfaces), at which an eective interfacial (or surface) tension acts. Surface tension has the units of force/length or equivalently energy/area, and so may be thought of as a negative surface pressure. Pressure is generally an isotropic force per area that acts through1

out the bulk of a uid: a small surface element dS will feel a total force p(x)dS owing to the local pressure eld p(x). If the surface S is closed, and the pressure uniform, the net pressure force acting on S is zero and the uid remains static. Pressure gradients correspond to body forces (with units of force per unit volume) within a uid, and so appear explicitly in the Navier-Stokes equations. Surface tension has the units of force per length, and its action is conned to the free surface. Consider for the sake of simplicity a perfectly at interface. A surface line element d will feel a total force d owing to the local surface tension (x). If the surface line element is a closed loop C , and the surface tension uniform, the net surface tension force acting on C is zero, and the uid remains static. If surface tension gradients arise, there may be a net force on the surface element that acts to distort it through driving ow. 1.2 Governing Equations The motion of a uid of uniform density and viscosity is governed by the Navier-Stokes equations, which represent a continuum statement of Newtons laws. u + u u = p + F + 2 u t u = 0. (1) (2)

This represents a system of 4 equations in 4 unknowns (the uid pressure p and the three components of the velocity eld u). Here F represents any body force acting on the uid; for example, in the presence of a gravitational eld, F = g, where g is the acceleration due to gravity. Surface tension acts only at the free surface; consequently, it does not appear in the NavierStokes equations, but rather enters through the boundary conditions. The boundary conditions appropriate at a uid-uid interface are formally developed in Lecture 2. We here simply state them for the simple case of a free surface (such as air-water, in which one of the uids is not dynamically signicant) in order to get a feeling for the scaling of surface tension. The normal stress balance at a free surface must be balanced by the curvature force associated with the surface tension: n T n = ( n) (3) where T = pI + [u + (u)T ] = pI + 2E is the stress tensor, E = 1/2 [u + (u)T ] is the deviatoric stress tensor, and n is the unit normal to the surface. The tangential stress at a free surface must balance the local surface tension gradient: n T t = t where t is the unit tangent to the interface. 1.3 The scaling of surface tension We consider a uid of density and viscosity = with a free surface characterized by a surface tension . The ow is marked by characteristic length- and velocity-scales of, respectively, a and U and evolves in the presence of a gravitational eld g = g z . We thus have a physical system dened in terms of six physical variables (, , , a, U, g ) that may be expressed in terms of three 2

(4)

fundamental units: mass, length and time. Buckinghams Theorem thus indicates that the system may be uniquely prescribed in terms of three dimensionless groups. We choose Re = Ua IN ERT IA = = Reynolds number V ISCOSIT Y U2 IN ERT IA = = Froude number ga GRAV IT Y (5)

Fr =

(6)

Bo =

GRAV IT Y ga2 = = Bond number CU RV AT U RE

(7)

The Reynolds number prescribes the relative magnitudes of inertial and viscous forces in the system, while the Froude number those of inertial and gravity forces. The Bond number indicates the relative importance of forces induced by gravity and surface tension. Note that these two forces are comparable when Bo = 1, which arises on a lengthscale corresponding to the capillary length: c = ( /(g ))1/2 . For an air-water surface, for example, 70 dynes/cm, = 1 g/cc and g = 980 cm/s2 , so that c 2mm. Bodies of water in air are dominated by the inuence of surface tension provided they are smaller than the capillary length. Roughly speaking, the capillary length prescribes the maximum size of pendant drops that may hang inverted from a ceiling, water-walking insects, and raindrops. Note that as a uid system becomes progressively smaller, the relative importance of surface tension and gravity increases; it is thus that surface tension eects are dominant in microscale engineering processes. Finally, we note that other frequently arising dimensionless group may be formed from the product of B , Re and F r: We = Ca = U 2 a IN ERT IA = = Weber number . CU RV AT U RE (8)

U V ISCOU S = = Capillary number . (9) CU RV AT U RE The Weber number indicates the relative magnitudes of inertial and curvature forces within a uid, and the capillary number those of viscous and curvature forces. Finally, we note that if the ow is marked by a Marangoni stress of characteristic magnitude /L, then an additional dimensionless group arises that characterizes the relative magnitude of Marangoni and curvature stresses: a /(L ). We now demonstrate how these dimensionless groups arising naturally from the nondimensionalization of Navier-Stokes equations and the surface boundary conditions. We rst introduce a dynamic pressure: pd = p g x, so that gravity appears only in the boundary conditions. We consider the special case of high Reynolds number ow, for which the characteristic dynamic pressure is U 2 . We dene dimensionless primed variables according to:
u = U u , pd = U 2 p d , x = ax , t =

a t . U

(10)

Nondimensionalizing the Navier-Stokes equations and appropriate boundary conditions yields the following system: u + u u t = p d + 3 1 2 u Re , u = 0. (11)

The normal stress condition assumes the dimensionless form: p d + 1 2 1 z + n E n = n Fr Re We (12)

The importance of surface tension relative to gravity and viscous stresses is prescribed by the relative magnitudes of the Weber, Froude and Reynolds numbers. In the high Re limit of interest, the normal force balance requires that the dynamic pressure be balanced by either gravitational or curvature stresses, the relative magnitudes of which are prescibed by the Bond number. The nondimensionalization scheme will depend on the physical system of interest. Our purpose here was simply to illustrate the manner in which the dimensionless groups arise in the theoretical formulation of the problem. Moreover, we see that those involving surface tension enter exclusively through the boundary conditions. References While this list of relevant textbooks is far from complete, we include it as a source of additional reading for the interested student. Rowlinson, J.S. and Widom, B., 1982, Molecular theory of capillarity, Dover. Israelachvili, J., 1995. Intermolecular and surface forces, 2nd Edition, Academic Press. DeGennes, P., Brochard-Wyart, M. and Quere, D., 2002. Perles, Gouttes, bulles, perles et ondes, Belin.

LECTURE 2: Stress Conditions at a Fluid-uid Interface We proceed by deriving the normal and tangential stress boundary conditions appropriate at a uid-uid interface characterized by an interfacial tension . Consider an interfacial surface S bound by a closed contour C (Figure 1). One may think of there being a force per unit length of magnitude in the s-direction at every point along C that acts to atten the surface S . Perform a force balance on a volume element V enclosing the interfacial surface S dened by the contour C : Du dV = Dt f dV + [t(n) + t( n)] dS + s d

Here indicates arclength and so d a length increment along the curve C . t(n) = n T is the stress vector, the force/area exerted by the upper (+) uid on the interface. The stress tensor is dened in terms of the local uid pressure and velocity eld as T = p I + [u + (u)T ]. = n T where T Similarly, the stress exerted on the interface by the lower (-) uid is t( n) = n T = p + (u ) ]. T I+ [u Physical interpretation of terms Du dV : inertial force associated with acceleration of uid within V Dt V V f dV : body forces acting on uid within V
S S C

t(n) dS : hydrodynamic force exerted at interface by uid +

t( n) dS : hydrodynamic force exerted at interface by uid s d : surface tension force exerted along perimeter C

Figure 1: A surface S and bounding contour C on an interface between two uids. The upper uid (+) has density and viscosity ; the lower uid (-), and . n represents the unit outward = n the unit inward normal. m the unit tangent to the contour C normal to the surface, and n and s the unit vector normal to C but tangent to S . 1

Figure 2: A Gaussian uid pillbox of height and radius spanning the interface evolves under the combined inuence of volume and surface forces. Now if is the typical lengthscale of the element V , then the acceleration and body forces will scale as 3 , but the surface forces will scale as 2 . Hence, in the limit of 0, we have that the surface forces must balance: [t(n) + t( n)] dS + s d = 0
S C

Now we have that t(n) = n T , = nT T t(n) = n

Moreover, the application of Stokes Theorem (see Appendix A) allows us to write s d = s n (s n) dS

where the tangential gradient operator, dened by s = [I nn] = n n

appears because and n are dened only on the surface. We proceed by dropping the subscript s on , with this understanding. The surface force balance thus becomes:

] dS = [n T n T

n ( n) dS

(1)

Now since the surface element is arbitrary, the integrand must vanish identically. One thus obtains the interfacial stress balance equation.

Figure 3: A denitional sketch of a uid-uid interface. Carats denote variables in the lower uid. Stress Balance Equation = n ( n) nT nT Interpretation of terms: n T: stress (force/area) exerted by + on - (will generally have both normal and tangential components) : stress (force/area) exerted by - on + (will generally have both normal and tangential comnT ponents) n ( n): normal curvature force per unit area associated with local curvature of interface, n. : tangential stress associated with gradients in surface tension. Both normal and tangential stress must be balanced at the interface. We consider each component in turn. Normal Stress Balance Taking n (2) yields the normal stress balance at the interface: n = ( n) nTn nT (3) (2)

The jump in normal stress across the interface must balance the curvature force per unit area. We note that a surface with non-zero curvature ( n = 0) reects a jump in normal stress across the interface. Tangential Stress Balance Taking t (2), where t is any unit vector tangent to the interface, yields the tangential stress 3

balance at the interface: t = t nTt nT Physical Interpretation: the LHS represents the jump in tangential components of the hydrodynamic stress at the interface the RHS represents the tangential stress associated with gradients in , as may result from gradients in temperature or chemical composition at the interface the LHS contains only velocity gradients, not pressure; therefore, a non-zero at a uid interface must always drive motion. (4)

Appendix A Recall Stokes Theorem: = Fd n ( F) dS

= m d, so that we have Along the contour C , d F m d = n ( F) dS

Now let F = f b, where b is an arbitrary constant vector. We thus have

(f b) m d =

n ( (f b)) dS

Now use standard vector identities to see: (f b) m = b (f m) (f b) = f ( b) b( f ) + b f f b = b( f ) + b f since b is a constant vector. We thus have 4

(f m) d = b

[n( f ) (f ) n] dS

Since b is arbitrary, we thus have (f m) d = [n( f ) (f ) n] dS

We now choose f = n, and recall that n m = s. One thus obtains s d = [n ( n) ( n) n] dS

= We note that

[n n + n( n) (n) n ] dS

n = 0 since must be tangent to the surface S , (n) n =

n) =

1 2 (1)

= 0 ,

and so obtain the desired result: s d = [ n ( n) ] dS

LECTURE 3: Fluid Statics We begin by considering static uid congurations, for which the stress tensor reduces to the form T = pI, so that n T n = p, and the normal stress balance assumes the form: p p = n

(1)

The pressure jump across the interface is balanced by the curvature force at the interface. Now since n T s = 0 for a static system , the tangential stress balance equation indicates that: 0 = . This leads us to the following important conclusion: There cannot be a static system in the presence of surface tension gradients. While pressure jumps can sustain normal stress jumps across a uid interface, they do not contribute to the tangential stress jump. Consequently, tangential surface stresses can only be balanced by viscous stresses associated with uid motion. We proceed by applying equation (1) to a number of static situations. 3.1 Stationary bubble We consider a spherical bubble of radius R submerged in a static uid. What is the pressure drop across the bubble surface? The curvature of the spherical surface is simply computed: 1 2 n = r = 2 (r2 ) = (2) r r R so the normal stress jump (1) indicates that 2 p p = . (3) R The pressure within the bubble is higher than that outside by an amount proportional to the surface tension, and inversely proportional to the bubble size. It is thus that small bubbles are louder than large ones when they burst at a free surface: champagne is louder than beer. We note that soap bubbles in air have two surfaces that dene the inner and outer surfaces of the soap lm; consequently, the pressure dierential is twice that across a single interface. 3.2 Static meniscus Consider a situation where the pressure within a static uid varies owing to the presence of a gravitational eld, p = p0 + gz , where p0 is the constant ambient pressure, and g = g z is the gravitational acceleration. The normal stress balance thus requires that the interface satisfy the Young-Laplace Equation: gz = n . (4) The vertical gradient in uid pressure must be balanced by the curvature pressure; as the gradient is constant, the curvature must likewise increase linearly with z. Such a situation arises in the static meniscus (see Figure 1).

The shape of the meniscus is prescribed by two factors: the contact angle between the air-water interface and the log, and the balance between hydrostatic pressure and curvature pressure. We 1

Figure 1: A denitional sketch of a planar meniscus at an air-water interface. The free surface is dened by z = (x), varying from its maximum elevation at its point of contact with the wall (x = 0) to zero at large x. The shape is prescribed by the Young-Laplace equation. treat the contact angle, , as given; it depends on the physics of the log-water-air interaction. The normal force balance is expressed by the Young-Laplace equation, where now = w a w is the density dierence between water and air. We dene the free surface by z = (x); equivalently, we dene a functional f (x, z ) = z (x) that vanishes on the surface. The normal to the surface is thus n = f z (x)x = |f | (1 + (x)2 )1/2 xx xx . 2 )3/2 (1 + x (5)

As deduced in Appendix A, the curvature of the free surface, n , may be expressed as n = (6)

Assuming that the slope of the meniscus remains small, x << 1, allows one to linearize equation (13), so that (11) assumes the form: g = xx . (7) Applying the boundary condition () = 0 and the contact condition x (0) = cot , and solving (16) thus yields: (x) = c cot()ex/c , (8) where c = g is the capillary length. The meniscus formed by a log in water is exponential, dying o on the scale of c .

Figure 2: A schematic illustration of the geometry of the circular hydraulic jump. A jet of radius a impacts a reservoir of outer depth H at a speed U . The curvature force associated with the surface tension depends only on the geometry of the jump; specically, on the radial distance R = R2 R1 and arclength s between the two nearest points up and downstream of the jump at which the surface z = h(r) has vanishing slope. 3.3 Radial force on a circular hydraulic jump Hydraulic jumps may be generated when a vertical jet strikes a at plate. The jet spreads radially, giving rise to a uid layer that generally thins with radius until reaching a critical radius at which it increases dramatically (see Figure 2 and http://www-math.mit.edu/ bush/jump.htm). We here calculate the total radial force acting on the jump surface owing to the curvature of the jump between points A and B, located at radii R1 and R2 , respectively. Assume that the points A and B are the points nearest the jump, respectively, upstream and downstream, at which the slope of the surface vanished identically. We know that the curvature force per unit area is ( n)n. We must integrate the radial component of this force over the jump surface: Fc = n (n r) dS (9)

Now dening the surface as z = h(r) allows us to express the area element as dS = r d(dr2 + dz 2 )1/2 = r d (1 + hr )1/2 dr , so that our radial force assumes the form 2 R2 Fc = n (n r) r (1 + hr )1/2 dr d .
0 R1

(10)

(11)

Now we use the appropriate forms for n and n from Appendix A, equation (41): n = z hr r 1 2 (1 + hr ) /2 , n = 3 rhr 1 d 1 /2 r dr (1 + h2 r) , (12)

so that ( n)(n r) =

1 d hr rhr 1/2 r dr (1 + h2 )1/2 (1 + h2 ) r r

(13)

and the radial curvature force becomes R2 hr 1 d rhr Fc = 2 r (1 + hr )1/2 dr 2 )1/2 r dr 2 )1/2 (1 + h (1 + h R1 r r R2 d rhr = 2 hr dr 1/2 dr (1 + h2 R1 r) b b rhr Integrating by parts, a udv = uv |b a a v du with u = hr , v = (1+h2 )1/2 , yields
r

(14) (15)

Fc = 2

rh2 r
R2

1 |R2 1/2 R1 (1 + h2 r) rhr hr r dr 1 /2 (1 + h2 r)

= 2

1/2 , yields Integrating by parts again, with u = r, v = (1 + h2 r) R2 2 1/2 R2 2 1/2 Fc = 2 r(1 + hr ) |R1 (1 + hr ) dr R1

R2

R1

rhr hr r dr 1/2 (1 + h2 r)

(16) (17)

R1

(18) (19)

since hr = 0 at r = R1 , R2 by assumption. We note also that R2 B 2 1/2 2 2 1/2 (1 + hr ) dr = (dr + dz ) =

= 2 (R2 R1 )

R2

(1 +

R1

1/2 h2 r)

dr

d = S ,

(20)

where S is dened as the total arclength of the surface between points A and B. We dene R = R1 R2 in order to obtain the simple result: Fc = 2 (S R) . (21) We note that this relation yields reasonable results in two limits of interest. First, for a at interface, S = R, so that Fc = 0. Second, for an abrupt jump in height of H , R = 0 and S = H , so that Fc = 2 H . This result is commensurate with the total force exerted by a cylindrical annulus of radius R and height H , which is deduced from the product of , the area 2 RH and the curvature 1/R: 1 Fc = (2 RH ) = 2 H (22) R The inuence of this force was found to be signicant for small hydraulic jumps. Its importance relative to the hydrostatic pressure in containing the jump is given by Curvature /R H 1 = = B Gravity g H R 0 where the Bond number is dened here as B0 = gH 2 4 . (24) (23)

The simple result (19) was derived by Bush & Aristo (2003), who also presented the results of an experimental study of circular hydraulic jumps. Their experiments indicated that the curvature force becomes appreciable for jumps with characteristic radius of less than 3 cm. While the inuence of surface tension on the radius of the circular hydraulic jump is generally small, it may have a qualitative inuence on the shape of the jump. In particular, it may prompt the axisymmetry-breaking instability responsible for the polygonal hydraulic jump structures discovered by Elegaard et al. (Nature, 1998), and more recently examined by Bush, Hosoi & Aristo (2004). See http://www-math.mit.edu/ bush/jump.htm. 3.4 Floating Bodies Floating bodies must be supported by some combination of buoyancy and curvature forces. Specifically, since the uid pressure beneath the interface is related to the atmospheric pressure P0 above the interface by p = P0 + gz + n , one may express the vertical force balance as Mg = z The buoyancy force Fb = z is thus sim bove the object and inside the line of tangency (Figure 3). A simple expression for the curvature force may be deduced using the rst of the Frenet-Serret equations (see Appendix B). Fc = z ( n)n d = z dt d = z (t1 t2 ) = 2 sin d (27) gz n d = gVb (26) p n d = Fb + Fc . (25)

At the interface, the buoyancy and curvature forces must balance precisely, so the Young-Laplace relation is satised: 0 = gz + n Integrating the uid pressure over the meniscus yields the vertical force balance: Fbm + Fcm = 0 . where Fbm = z gz n d = gVm 5 (30) (29) (28)

Cm

tx g

LC

Vm n Cm

air water

tx

t1 C

r t2

Figure 3: A non-wetting two-dimensional body of radius r and mass M oats on a free surface with surface tension . In general, its weight M g must be supported by some combination of curvature and buoyancy forces. Vb and Vm denote the uid volumes displaced, respectively, inside and outside the line of tangency.

Fcm

= z

Cm

( n)n d = z

Cm

dt d = z (tx t2 ) = 2 sin d

(31)

where we have again used the Frenet-Serret equations to evaluate the curvature force. Equations (27)-(31) thus indicate that the curvature force acting on the oating body is expressible in terms of the uid volume displaced outside the line of tangency: Fc = gVm . (32)

The relative magnitude of the buoyancy and curvature forces supporting a oating, non-wetting body is thus prescribed by the relative magnitudes of the volumes of the uid displaced inside and outside the line of tangency: Fb Vb = Fc Vm . (33)

For 2D bodies, we note that since the meniscus will have a length comparable to the capillary length, c = ( /(g ))1/2 , the relative magnitudes of the buoyancy and curvature forces, r Fb Fc c , (34)

is prescribed by the relative magnitudes of the body size and capillary length. Very small oating objects (r c ) are supported principally by curvature rather than buoyancy forces. This result has been extended to three-dimensional oating objects by Keller (1998, Phys. Fluids, 10, 3009-3010). Water-walking Insects

Figure 4: Water-walking insects deect the free surface, thus generating curvature forces that bear their weight. The water strider has characteristic length 1cm and weight 1-10 mg. See http://wwwmath.mit.edu/ dhu/Striderweb/striderweb.html .

Figure 5: The water strider legs are covered with hair, rendering them eectively non-wetting. The tarsal segment of its legs rest on the free surface. The free surface makes an angle with the horizontal, resulting in an upward curvature force per unit length 2 sin that bears the insects weight.

Figure 6: The relation between the maximum curvature force Fs = P and body weight Fg = M g for 342 species of water striders. P = 2(L1 + L2 + L3 ) is the combined lengths of the tarsal segments (see strider B). From Hu, Chan & Bush (Nature, 424, 2003). Small objects such as paper clips, pins or insects may reside at rest on a free surface provided the curvature force induced by their deection of the free surface is sucient to bear their weight. For example, for a body of contact length L and total mass M , static equilibrium on the free surface requires that: Mg < 1 (35) 2 Lsin where is the angle of tangency of the oating body. This simple criterion is an important geometric constraint on water-walking insects. Figure 6 indicates the dependence of contact length on body weight for over 300 species of water-striders, the most common water walking insect. Note that the solid line corresponds to the requirement (35) for static equilibrium. Smaller insects maintain a considerable margin of safety, while the larger striders live close to the edge. The size of water-walking insects is limited by the constraint (35).

If body proportions were independent of size L, one would expect the body weight to scale as L3 1/3 and the curvature force as L. Isometry would thus suggest a dependence of the form Fc Fg , represented as the dashed line in Figure 6. The fact that the best t line has a slope considerably larger than 1/3 indicates a variance from isometry: the legs of large water striders are proportionally longer.

Appendix A: Computing curvatures We see the appearance of the divergence of the surface normal, n, in the normal stress balance. We proceed by briey reviewing how to formulate this curvature term in two common geometries. In cartesian coordinates (x, y, z ), we consider a surface dened by z = h(x, y ). We dene a functional f (x, y, z ) = z h(x, y ) that necessarily vanishes on the surface. The normal to the surface is dened by z hx x hy y f n = (36) = 1 2 2 |f | (1 + hx + hy ) /2
2 (hxx + hyy ) (hxx h2 y + hyy hx ) + 2hx hy hxy 2 3/2 (1 + h2 x + hy )

and the local curvature may thus be computed: n =

(37)

In the simple case of a two-dimensional interface, z = h(x), these results assume the simple forms: n = z hx x 2 (1 + hx )1/2 , n = hxx 3/2 (1 + h2 x) . (38)

Note that n is dimensionless, while n has the units of 1/L. In 3D polar coordinates (r, , z ), we consider a surface dened by z = h(r, ). We dene a functional g (r, , z ) = z h(r, ) that vanishes on the surface, and compute the normal: n = z hr r 1 g r h = 1 2 1 /2 |g | (1 + h2 r + r 2 h ) , (39)

from which the local curvature is computed: n =

(40)

In the case of an axisymmetric interface, z = h(r), these results reduce to: n = z hr r 2 (1 + hr ) , n = rhr r2 hrr 3/2 r2 (1 + h2 r) . (41)

Appendix B: Frenet-Serret Equations

Dierential geometry yields relations that are often useful in computing curvature forces on 2D interfaces. ( n) n = ( n) t = dt d dn d (42)

(43)

Note that the LHS of the rst of (42) is proportional to the curvature pressure acting on an interface.

10

LECTURE 4: Marangoni ows Marangoni ows are those driven by surface tension gradients. In general, surface tension depends on both the temperature and chemical composition at the interface; consequently, Marangoni ows may be generated by gradients in either temperature or chemical concentration at an interface. In lecture 2, we derived the tangential stress balance at a free surface: n T t = t , (1)

where n is the unit outward normal to the surface, and t is any unit tangent vector. The tangential component of the hydrodynamic stress at the surface must balance the tangential stress associated with gradients in . Such Marangoni stresses may result from gradients in temperature or chemical composition at the interface. For a static system, since n T t = 0, the tangential stress balance equation indicates that: 0 = . This leads us to the following important conclusion:

There cannot be a static system in the presence of surface tension gradients. While pressure jumps can sustain normal stress jumps across a uid interface, they do not contribute to the tangential stress jump. Consequently, tangential surface stresses can only be balanced by viscous stresses associated with uid motion. 4.1 Tears of wine The rst Marangoni ow considered was the tears of wine phenomenon (Thomson 1885), which actually predates Marangonis rst published work on the subject by a decade. The tears of wine phenomenon is readily observed in a glass of any but the weakest wines following the establishment of a thin layer of wine on the walls of the wine glass. The tears or legs of wine are taken by sommeliers to be an indicator of the quality of wine. An illustration of the tears of wine phenomenon is presented in Figure 1 (see also http://wwwmath.mit.edu/ bush/tears.html ). Evaporation of alcohol occurs everywhere along the free surface. The alcohol concentration in the thin layer is thus reduced relative to that in the bulk owing to the enhanced surface area to volume ratio. As surface tension decreases with alcohol concentration, the surface tension is higher in the thin lm than the bulk; the associated Marangoni stress drives upow throughout the thin lm. The wine climbs until reaching the top of the lm, where it accumulates in a band of uid that thickens until eventually becoming gravitationally unstable and releasing the tears of wine. The tears or legs roll back to replenish the bulk reservoir, but with uid that is depleted in alcohol. The ow relies on the transfer of chemical potential energy to kinetic and ultimately gravitational potential energy. The process continues until the alcohol is completely evaporated. 4.2 Surfactants Surfactants are molecules that have an anity for interfaces; common examples include soap and oil. Owing to their molecular structure (often a hydrophylic head and hydrophobic tail), they nd it energetically favourable to reside at the free surface. Their presence reduces the surface tension; consequently, gradients in surfactant concentration result in surface tension gradients. Surfactants thus generate a special class of Marangoni ows. There are many dierent types of surfactants, some of which are insoluble (and so remain on the surface), others of which are soluble 1

Figure 1: The tears of wine. Fluid is drawn from the bulk up the thin lm adjoining the walls of the glass by Marangoni stresses induced by evaporation of alcohol from the free surface. in the suspending uid and so diuse into the bulk. For a wide range of common surfactants, surface tension is a monotonically decreasing function of until a critical concentration is achieved, beyond which remains constant. The concentration of surfactant on a free surface evolves according to + s (us ) + (s n)(u n) = J (, Cs ) + Ds 2 s , t (2)

where us is the surface velocity, s is the surface gradient operator and Ds is the surface diusivity of the surfactant. J is a surfactant source term associated with adsorption onto or desorption from the surface, and depends on both the surface surfactant concentration and the concentration in the bulk Cs . Tracing the evolution of a contaminated free surface requires the use of NavierStokes equations, relevant boundary conditions and the surfactant evolution equation (2). The dependence of surface tension on surfactant concentration, (), requires the coupling of the ow eld and surfactant eld. In certain special cases, the system may be made more tractable. For example, for insoluble surfactants, J = 0, and many surfactants have suciently small Ds that surface diusivity may be safely neglected. The principal dynamical inuence of surfactants is to impart an eective elasticity to the interface. Specically, the presence of surfactants will serve not only to alter the normal stress balance (through the reduction of ), but also the tangential stress balance (through the generation of Marangoni stresses). The presence of surfactants will act to suppress any uid motion characterized by non-zero surface divergence. For example, consider a uid motion characterized by a radially divergent surface motion. The presence of surfactants results in the redistribution of surfactants: is reduced near the point of divergence. The resulting Marangoni stresses act to suppress the surface motion, resisting it through an eective surface elasticity. Similarly, if the ow is characterized by a radial convergence, the resulting accumulation of surfactant in the region of 2

Figure 2: The footprint of a whale, caused by the whales sweeping biomaterial to the sea surface. The biomaterial acts as a surfactant in locally suppressing the capillary waves evident elsewhere on the sea surface. Observed in the wake of a whale on a whale watch from Boston Harbour. convergence will result in Marangoni stresses that serve to resist it. It is this eective elasticity that gives soap lms their longevity: the divergent motions that would cause a pure liquid lm to rupture are suppressed by the surfactant layer on the soap lm surface. The ability of surfactant to suppress ows with non-zero surface divergence is evident throughout the natural world. It was rst remarked upon by Pliny the Elder, who rationalized that the absence of capillary waves in the wake of ships is due to the ships stirring up surfactant. This phenomenon was also widely known to spear-shermen, who poured oil on the water to increase their ability to see their prey, and by sailors, who would do similarly in an attempt to calm troubled seas. Finally, the suppression of capillary waves by surfactant is responsible for the footprints of whales (see Figure 2). In the wake of whales, even in turbulent roiling seas, one seas patches on the sea surface (of characteristic width 5-10m) that are perfectly at. These are generally acknowledged to result from the whales sweeping biomaterial to the surface with their tails; this biomaterial serves as a surfactant that suppresses capillary waves. 4.3 The soap boat Consider a oating body with perimeter C in contact with the free surface, which we assume for the sake of simplicity to be at. Recalling that may be thought of as a force per length in a direction tangent to the surface, we see that the total surface tension force acting on the body is: Fc = s d , (3)
C

where s is the unit vector tangent to the free surface and normal to C , and d is an incremental arc length along C . If is everywhere constant, then this line integral vanishes identically by the Divergence Theorem. However, if = (x), then it may result in a net propulsive force. 3

Figure 3: The soap boat. A oating body (length 2.5cm) contains a small volume of soap, which serves as its fuel in propelling it across the free surface.The soap exits the rear of the boat, decreasing the local surface tension. The resulting fore-to-aft surface tension gradient propels the boat forward. The water surface is covered with Thymol blue, which parts owing to the presence of soap, which is visible as a white streak. The soap boat may be simply generated by coating one end of a toothpick with soap, which acts to reduce surface tension (see Figure 3). The concomitant gradient in surface tension results in a net propulsive force that drives the boat away from the soap. We note that an analogous Marangoni propulsion technique arises in the natural world: certain water-walking insects eject surfactant and use the resulting surface tension gradients as an emergency mechanism for avoiding predation. Moreover, when a pine needle falls into a lake or pond, it is propelled across the surface in an analogous fashion owing to the inuence of the resin at its base decreasing the local surface tension. 4.4 Bubble motion Theoretical predictions for the rise speed of small drops or bubbles do not adequately describe observations. Specically, air bubbles rise at low Reynolds number at rates appropriate for rigid spheres with equivalent buoyancy in all but the most carefully cleaned uids. This discrepancy may be rationalized through consideration of the inuence of surfactants on the surface dynamics. The ow generated by a clean spherical bubble of radius a rising at low Re = U a/ is intuitively obvious. The interior ow is toroidal, while the surface motion is characterized by divergence and convergence at, respectively, the leading and trailing surfaces. The presence of surface contamination changes the ow qualitatively. The eective surface elasticity imparted by the surfactants acts to suppress the surface motion. Surfactant is generaly swept to the trailing edge of the bubble, where it accumulates, giving rise to a local decrease in surface tension. The resulting fore-to-aft surface tension gradient results in a 4

Marangoni stress that resists surface motion, and so rigidies the bubble surface. The air bubble thus moves as if its surface were stagnant, and it is thus that its rise speed is commensurate with that predicted for a rigid sphere: the no-slip boundary condition is more appropriate than the free-slip. Finally, we note that the characteristic Marangoni stress /a is most pronounced for small bubbles. It is thus that the inuence of surfactants is greatest on small bubbles.

LECTURE 5: Fluid jets We consider here the form and stability of uid jets falling under the inuence of gravity. 5.1 The shape of a falling uid jet Consider a circular orice of radius a ejecting a ux Q of uid of density and kinematic viscosity (Figure 1). The resulting jet is shot downwards, and accelerates under the inuence of gravity g z . We assume that the jet Reynolds number Re = Q/(a ) is suciently high that the inuence of viscosity is negligible; furthermore, we assume that the jet speed is independent of radius, and so adequately described by U (z ). We proceed by deducing the shape r(z ) and speed U (z ) of the evolving jet. Applying Bernoullis Theorem at points A and B: 1 2 1 U + gz + PA = U 2 (z ) + PB 2 0 2 (1)

The local curvature of slender threads may be expressed in terms of the two principal radii of curvature, R1 and R2 : n = 1 1 1 + R1 R2 r

Thus, the uid pressures within the jet at points A and B may be simply related to that of the ambient, P0 : PA P0 + Substituting into (1) thus yields 1 2 1 U0 + gz + P0 + = U 2 (z ) + P0 + 2 a 2 r from which one nds U (z ) = U0 2 z 2 a 1/ 2 1+ + 1 , Fr a We r (4) (3) a , PB P0 + r (2)

where we dene the dimensionless groups: Fr =

2 U0 INERTIA = = Froude Number , ga GRAVITY

(5)

We =

2a INERTIA U0 = = Weber Number , CURVATURE

(6)

Figure 1: A uid jet extruded from an orice of radius a accelerates under the inuence of gravity. Its shape is inuenced both by the gravitational accelerationg and the surface tension .

Now ux conservation requires that

Q = 2 from which one obtains r(z ) = a

U (z )r(z ) dr = a2 U0 = r2 U (z )

(7)

U0 U (z )

1/2

2 z 2 a 1/4 1+ + 1 Fr a We r

(8)

This may be solved algebraically to yield the thread shape r(z )/a, then this result substituted into (4) to deduce the velocity prole U (z ). In the limit of W e , one obtains r = a 2gz 1+ 2 U0 1/4 , U (z ) = U0 2gz 1+ 2 U0 1 / 2

5.2 The Plateau-Rayleigh Instability We here summarize the work of Plateau and Rayleigh on the instability of cylindrical uid jets bound by surface tension. It is precisely this Rayleigh-Plateau instability that is responsible for the pinch-o of thin water jets emerging from kitchen taps (see Figure 2). 2

Figure 2: The capillary-driven instability of a water thread falling under the inuence of gravity. The initial jet diameter is approximately 3 mm. The equilibrium base state consists of an innitely long quiescent cylindrical inviscid uid column of radius R0 , density and surface tension (Figure 3). The inuence of gravity is neglected. The pressure p0 is constant inside the column and may be calculated by balancing the normal stresses with surface tension at the boundary. Assuming zero external pressure yields . (9) p0 = n p0 = R0 We consider the evolution of innitesimal varicose perturbations on the interface, which enables us to linearize the governing equations. The perturbed columnar surface takes the form: = R0 + et+ikz , R (10)

where the perturbation amplitude R0 , is the growth rate of the instability and k is the wave number of the disturbance in the z -direction. The corresponding wavelength of the varicose perturbations is necessarily 2 /k . We denote by u r the radial component of the perturbation velocity, u z the axial component, and p the perturbation pressure. Substituting these perturbation elds into the Navier-Stokes equations and retaining terms only to order yields: 1 p u z = . t z The linearized continuity equation becomes: u r u r + +u z = 0 . r r 3 u r 1 p = t r (11) (12)

(13)

n , p
0

, p + ~ p
0

Ro+

Ro I. Steady State II. Perturbed State

Figure 3: A cylindrical column of initial radius R0 is comprised of uid of inviscid uid of density and bound by surface tension . We anticipate that the disturbances in velocity and pressure will have the same form as the surface disturbance (10), and so write the perturbation velocities and pressure as:
t+ikz t+ikz u , u and p = P (r)et+ikz . r = R(r )e z = Z (r )e

(14)

Substituting (14) into equations (11) through (13) yields the linearized equations governing the perturbation elds: Momentum equations: R = 1 dP dr ik P

(15) (16)

Z = Continuity:

dR R + + ikZ = 0 . dr r Eliminating Z (r) and P (r) yields a dierential equation for R(r): r2 d2 R dR 2 + r 1 + ( kr ) R=0 . dr2 dr

(17)

(18)

This corresponds to modied Bessel Equation of order 1, whose solutions may be written in terms of the modied Bessel functions of the rst and second kind, respectively, I1 (kr) and K1 (kr). We note that K1 (kr) as r 0; therefore, the well-behavedness of our solution requires that R(r) take the form R(r) = CI1 (kr) , (19) 4

0.34( / R3 )
0

0.2

0.4

kR0

0.6

0.8

Figure 4: The dependence of the growth rate on the wavenumber k for the Rayleigh-Plateau instability. where C is an as yet unspecied constant to be determined later by application of appropriate boundary conditions. The pressure may be obtained from (19) and (15), and by using the Bessel function identity ( ) = I ( ): I0 1 C P (r) = I0 (kr) . (20) k We proceed by applying appropriate boundary conditions. The rst is the kinematic condition on the free surface: R r . (21) =u n u t Substitution of (19) into this condition yields C= . I1 (kR0 ) (22)

Second, we require a normal stress balance on the free surface: p0 + p = n . (23) 1 +R , where R1 and R2 are the principal radii of curvature 2 (24) (25)

1 We write the curvature as n = R 1 of the jet surface: 1 1 1 = 2 et+ikz t + ikz R1 R0 R0 R0 + e 1 = k 2 et+ikz . R2 Substitution of (24) and (25) into equation (23) yields: t+ikz 2 p0 + p = 2 1 k 2 R0 e . R 0 R0

(26)

Figure 5: The eld of stationary capillary waves excited on the base of a water jet impinging on a horizontal water reservoir. The grid at right is millimetric. Cancellation via (9) yields the equation for p accurate to order : p = t+ikz 2 1 k 2 R0 e . 2 R0

(27)

Combining (20), (22) and (27) yields the dispersion relation, that indicates the dependence of the growth rate on the wavenumber k : 2 =
I1 (kR0 ) 3 kR0 I (kR ) R0 0 0

2 1 k 2 R0

(28)

We rst note that unstable modes are only possible when kR0 < 1 (29)

The column is thus unstable to disturbances whose wavelengths exceed the circumference of the cylinder. A plot for the dispersion relation is shown in Figure 4. 6

The fastest growing mode occurs for kR0 = 0.697, i.e. when the wavelength of the disturbance is max 9.02R0 . (30)

By inverting the maximum growth rate max one may estimate the characteristic break up time: 3 R0 tbreakup 2.91 . (31) A water jet of diameter 1cm has a characteristic break-up time of about 1/8 s, which is consistent with casual observation of jet break-up in a kitchen sink. When a vertical water jet impinges on a horizontal reservoir of water, a eld of standing waves may be excited on the base of the jet (see Figure 5). The wavelength is determined by the requirement that the wave speed correspond to the local jet speed: U = /k . Using our dispersion relation (28) thus yields 2 I1 (kR0 ) 2 2 U2 = 2 = 1 k R . (32) 0 2 I (kR ) k kR0 0 0 Provided the jet speed U is known, this equation may be solved in order to deduce the wavelength of the waves that will travel at U and so appear to be stationary in the lab frame. For jets falling from a nozzle, the result (4) may be used to deduce the local jet speed. 5.3 Fluid Pipes (see http://www-math.mit.edu/ bush/pipes.html) The following system may be readily observed in a kitchen sink. When the volume ux exiting the tap is such that the falling stream has a diameter of 2-3mm, obstructing the stream with a nger at a distance of several centimeters from the tap gives rise to a stationary eld of varicose capillary waves upstream of the nger. If the nger is dipped in liquid detergent (soap) before insertion into the stream, the capillary waves begin at some critical distance above the nger, below which the stream is cylindrical. Closer inspection reveals that the surface of the jets cylindrical base is quiescent. An analogous phenomenon arises when a vertical uid jet impinges on a deep water reservoir (Figures 5 and 6). When the reservoir is contaminated by surfactant, the surface tension of the reservoir is diminished relative to that of the jet. The associated surface tension gradient draws surfactant a nite distance up the jet, prompting two salient alterations in the jet surface. First, the surfactant suppresses surface waves, so that the base of the jet surface assumes a cylindrical form (Figure 6). Second, the jet surface at its base becomes stagnant: the Marangoni stresses associated with the surfactant gradient are balanced by the viscous stresses generated within the jet. The quiescence of the jet surface may be simply demonstrated by sprinkling a small amount of talc or lycopodium powder onto the jet. The uid jet thus enters a contaminated reservoir as if through a rigid pipe. A detailed theoretical description of the uid pipe is given in Hancock & Bush (JFM, 466, 285304). We here present a simple scaling that yields the dependence of the vertical extent H of the uid pipe on the governing system parameters. We assume that, once the jet enters the uid pipe, a boundary layer develops on its outer wall owing to the no-slip boundary condition appropriate there. Balancing viscous and Marangoni stresses on the pipe surface yields V , H H 7 (33)

Figure 6: The uid pipe generated by a falling water jet impinging on a contaminated water reservoir. The eld of stationary capillary waves is excited above the uid pipe. The grid at right is millimetric. where is the surface tension dierential between the jet and reservoir, V is the jet speed at the top of the uid pipe, and H is the boundary layer thickness at the base of the uid pipe. We assume that the boundary layer thickness increases with distance z from the inlet according to classical boundary layer scaling: z 1/2 . (34) a a2 V Substituting for (H ) from (34) into (33) yields H ( )2 . V 3 (35)

The pipe height increases with the surface tension dierential and pipe radius, and decreases with uid viscosity and jet speed.

LECTURE 6: Fluid Sheets

The dynamics of high-speed uid sheets was rst considered by Savart after his early work on electromagnetism with Biot, and was subsequently examined in a series of papers by Taylor (Proc. Roy. Soc., 1959). They have recently received a great deal of attention owing to their relevance in a number of spray atomization processes. Such sheets may be generated from a variety of source conditions, for example, the collision of jets on rigid impactors, and jet-jet collisions. There is generally a curvature force acting on the sheet edge which acts to contain the uid sheet. For a 2D (planar) sheet, the magnitude of this curvature force is given by Fc = ( n) n d , (1)
C

Using the rst Frenet-Serret equation (Lecture 2, Appendix B), ( n) n = thus yields Fc = dt , d (2)

dt d = (t1 t2 ) = 2 x. d

(3)

There is thus an eective force per unit length 2 along the length of the sheet rim acting to contain the rim. We now consider how this result may be applied to compute sheet shapes for three distinct cases: i) a circular sheet, ii) a lenticular sheet with unstable rims, and iii) a lenticular sheet with stable rims. 6.1 Circular Sheet We consider the geometry considered in Savarts original experiment. A vertical uid jet strikes a small horizontal circular impactor. If the ow rate is suciently high that gravity does not inuence the sheet shape, the uid is ejected radially, giving rise to a circular free uid sheet (Figure 1). The sheet radius is prescribed by a balance of radial forces; specically, the inertial force must balance the curvature force: u2 h = 2 (4)

Continuity requires that the sheet thickness h depend on the speed u, jet ux Q and radius r as h = Q 2 ru . (5)

Experiments (specically, tracking of particles suspended within the sheet) indicate that the sheet speed u is independent of radius; consequently, the sheet thickness decreases as 1/r. Substituting the form (5) for h into the force balance (4) yields the sheet radius, or so-called Taylor radius: RT = 1 Qu 4 (6)

Figure 1: A circular uid sheet generated by the impact of a water jet on a circular impactor (below). The impacting circle has a diameter of 1 cm. The sheet radius increases with source ux and sheet speed, but decreases with surface tension. We note that the uid proceeds radially to the sheet edge, where it accumulates until going unstable via a modied Rayleigh-Plateau instability, often referred to as the Rayleigh-Plateau-Savart instability, as it was rst observed on a sheet edge by Savart. 6.2 Lenticular sheets with unstable rims We now consider the non-axisymmetric uid sheet, such as may be formed by the oblique collision of water jets (see Figure 2), a geometry originally considered by Taylor (1960). Fluid is ejected radially from the origin into a sheet with ux distribution given by Q(), so that the volume ux owing into the sector between and + d is Q () d. As in the previous case of the circular sheet, the sheet rims are unstable, and uid drops are continuously ejected therefrom. The sheet shape is computed in a similar manner, but now depends explicitly on the ux distribution within the sheet, Q(). The normal force balance on the sheet edge now depends on the normal component of the sheet speed, un : u2 (7) n h = 2 . The sheet thickness is again prescribed by (5), but now Q = Q(), so the sheet radius R() is given by u2 n Q( ) R() = (8) 4 u Computing sheet shapes thus relies on measurement of the ux distribution Q() within the sheet.

Figure 2: A water sheet generated by the collision of water jets at left. The uid streams radially outward in a thinning sheet; once the uid reaches the sheet rim, it is ejected radially in the form of droplets. From G.I. Taylor (1960). 6.3 Lenticular sheets with stable rims In a certain region of parameter space, specically, with uids more viscous than water, one may encounter uid sheets with stable rims (see http://www-math.mit.edu/ bush/bones.html). The force balance describing the sheet shape must change accordingly. When rims are stable, uid entering the rim proceeds along the rim. As a result, there is a centripetal force normal to the uid rim associated with ow along the curved rim that must be considered in order to correctly predict the sheet shapes. The relevant geometry is presented in Figure 3. r() is dened to be the distance from the origin to the rim centreline, and un () and ut () the normal and tangential components of the uid velocity in the sheet where it contacts the rim. v () is dened to be the velocity of ow in the rim, R() the rim radius, and () the angle between the position vector r and the local tangent to the rim centreline. Finally, rc () is dened to be the radius of curvature of the rim centreline, and s the arc length along the rim centreline. The dierential equations governing the shape of a stable uid rim bounding a uid sheet may be deduced by consideration of conservation of mass in the rim and the local normal and tangential force balances at the rim. For a steady sheet, continuity requires that the volume ux from the sheet balance the tangential gradient in volume ux along the rim: 0 = un h v R2 s . (9)

The normal force balance requires that the curvature force associated with the rims surface tension balance the force resulting from the normal ow into the rim from the uid sheet and the centripetal force resulting from the ow along the curved rim: u2 nh + R2 v 2 = 2 . rc 3 (10)

Rim Centreline Rim

ut

! un

uo r (") "

uo v

Figure 3: A schematic illustration of a uid sheet bound by stable rims. Note that the force balance (7) is here augmented by the centripetal force. The tangential force balance at the rim requires a balance between tangential gradients in tangential momentum ux, tangential gradients in curvature pressure, viscous resistance to stretching of the rim, and the tangential momentum ux arriving from the sheet. For most applications involving high-speed sheets, the Reynolds number characterizing the rim dynamics is suciently large that viscous resistance may be safely neglected. Moreover, the curvature term n generally depends on ; however, accurate to O(R/rc ), we may use n = 1/R. One thus obtains: R2 1 2 2 ( R v ) = hut un . (11) s s R Equations (9)-(11) must be supplemented by the continuity relation, h(r, ) = Q() , u0 r (12)

in addition to a number of relations that follow directly from the system geometry: un = u0 sin , ut = u0 cos sin 1 = +1 . rc r (13) (14)

This resulting system of equations may be nondimensionalized, and reduce to a set of coupled ordinary equations in the four variables r(), v (), R() and (). Given a ux distribution, Q(), 4

Figure 4: A water bell produced by the impact of a descending water jet and a solid impactor. The impactor radius is 1 cm. Fluid is splayed radially by the impact, then sags under the inuence of gravity. The sheet may close on itself owing to the azimuthal curvature of the bell. the system may be integrated to deduce the sheet shape. 6.4 Water bells All of the uid sheets considered thus far have been conned to a plane. In 6.1, we considered circular sheets generated from a vertical jet striking a horizontal impactor. The sheet remains planar only if the ow is suciently fast that the uid reaches its Taylor radius before sagging substantially under the inuence of gravity. Decreasing the ow rate will make this sagging more pronounced, and the sheet will no longer be planar. While one might expect the sheet to fall along a parabolic trajectory, the toroidal curvature of the bell induces curvature pressures that act to close the sheet. Consequently, the sheet may actually close upon itself, giving rise to a water bell, as illustrated in Figure 4. We proceed by outlining the theory required to compute the shapes of water bells. We consider a uid sheet extruded radially at a speed u0 and subsequently sagging under the inuence of a gravitational eld g = g z . The inner and outer sheet surfaces are characterized by a constant surface tension . The sheet has constant density , thickness t(r, z ) and is assumed to be inviscid. Q is the total volume ux in the sheet. We dene the origin to be the center of the impact plate; r and z are, respectively, the radial and vertical distances from the origin. u is the sheet speed, and the angle made between the sheet and the vertical. rc is the local radius of curvature of a meridional line, and s the arc length along a meridional line measured from the origin. Finally, P is the pressure dierence between the outside and inside of the bell. 5

Flux conservation requires that Q = 2 rtu while Bernoullis Theorem indicates that u2 = u2 0 + 2gz The total curvature force acting normal to the bell surface is given by 1 cos 2 n = 2 + . rc r (16) , (15)

(17)

Note that the factor of two results from there being two free surfaces. Therefore, the force balance normal to the sheet takes the form: 2 2 cos tu2 + P + gt sin = 0 rc r rc (18)

Equations (15), (16) and (18) may be appropriately nondimensionalized and integrated to determine the shape of the bell.