Instability of Liquid Surfaces and the Formation of DropsJoseph B. Keller and Ignace Kolodner Citation: Journal of Applied Physics 25, 918 (1954); doi: 10.1063/1.1721770 View online: http://dx.doi.org/10.1063/1.1721770 View Table of Contents: http://scitation.aip.org/content/aip/journal/jap/25/7?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Drop formation in a falling stream of liquid Am. J. Phys. 73, 415 (2005); 10.1119/1.1866100 Instabilities and drop formation in cylindrical liquid jets in reduced gravity Phys. Fluids 14, 3432 (2002); 10.1063/1.1501825 Instabilities of Two Liquid Drops in Contact Comput. Sci. Eng. 3, 16 (2001); 10.1109/5992.947104 Instability of a deformed liquid drop in an acoustic field Phys. Fluids 7, 2601 (1995); 10.1063/1.868708 Static shape and instability of an acoustically levitated liquid drop Phys. Fluids A 3, 2497 (1991); 10.1063/1.858192

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.22.67.7 On: Tue, 02 Dec 2014 15:06:22

918 EARL L. STEELE

The evaluation of this expression involves considerable numerical work since the number of terms which must be included for accuracy depends upon the value of (Lo/ L).

These results have been plotted for several values of the ratio (Lo/ L) and are shown in Fig. 3. It can be seen that the sharp gradient introduced at the junction by suddenly reducing the diode voltage to zero causes the holes to surge out, resulting in a large back current,

ID

:.­~o. •

~'''''' -

i If o .

i rx:-

• / ........

/ /

.4 I v-<.'---I

I " "-- tt .. • 0_0'

'v'----

¥., ,

.... r'fr.O.OI , ,

~ " .... --K .... ~ 0

0.2 0.4 0.6 0., 1.0 R[\.&TIVE DlffANCE THROUGH DtOOf IX'LJ

FIG. 4. Graph showing hole density variation through ger­manium for different time intervals after removal of forward bias voltage on diode. For the case shown, it was assumed that Lo/ L= 1.

JOURNAL OF APPLIED PHYSICS

"the spike," which falls off very rapidly as the sharp gradient is decreased. The decay of the current is noted to be extermely rapid for small ratios of (Lo/ L).

For longer times the current becomes essentially exponential for large ratios of (Lo/ L) but deviates con­siderably from the simple exponential for (t/ T<O.l).

The change of hole density distribution with time in the germanium for one value of the ratio (Lo/ L= 1) is shown in Fig. 4. The sharp gradient mentioned before is obvious from this figure.

4. CONCLUSIONS

The important factors contributing to an improve­ment in frequency response and a corresponding reduc­tion in spike current appear from the foregoing analysis. A reduction of the lifetime of holes in the bulk ger­manium as well as the use of thinner pellets will both tend to reduce the spike. The important ratio (Lo/ L) is also affected (increased) by a decrease in the lifetime because of the effect on the "diffusion length." This can be compensated for, however, by simply reducing the pellet thickness (Lo).

VOLUME 25. NUMBER 7 JULY, 1954

Instability of Liquid Surfaces and the Formation of Drops

JOSEPH B. KELLER AND IGNACE KOLODNER

Institute of Mathematical Sciences, New York University, New York, New York (Received October 19, 1953)

The theory of Taylor instability is extended to take account of surface tension, and the most unstable mode is determined. The resulting theory is then used to estimate the size of drops which might be formed as a result of this instability.

I. INTRODUCTION

I T has long been known that certain flows of liquids with free boundary surfaces are unstable. For steady

flows this means that a small perturbation introduced at a point of the flow produces a disturbance which in­creases in the flow direction until it completely alters the motion. This mechanism was used by Rayleigh to describe the breakup of a steady jet of liquid into drops. In the case of unsteady flows, instability implies that certain initial perturbations will increase in time until they significantly change the flow. Taylorl showed that the plane free boundary of a semi-infinite liquid is un­stable if the liquid moves with constant acceleration normal to the surface, with the acceleration directed into the liquid. Previously, Penney and Price2 had shown that the surface of a pUlsating underwater ex­plosion bubble is likewise unstable. In this case, also,

1 G. I. Taylor, Proc. Roy. Soc. (London) A201, 192 (1950). 2 W. G. Penney and A. T. Price, British Report SW-27 (1942).

the fluid moves with accelerated motion normal to the surface.

With these results in mind, we wish to account for the formation of drops resulting from the breakup of accelerated plane, cylindrical and spherical layers of liquid. These phenomena occur whenever explosives are detonated in or near liquids. For example, the spray dome produced by an underwater explosion results from a layer of water thrown into the air by the shock wave from the explosion. The present theory attempts to describe the breakup of this layer into drops of spray. As a second example, consider the production of an aerosol (i.e., dispersion of liquid drops in air) by a bomb containing an explosive surrounded by a layer of liquid. The high pressure gases resulting from the explosion propel the liquid layer outward, and it ultimately breaks up into drops. The present theory also attempts to account for this breakup.

In our proposed mechanism of breakup we assume

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.22.67.7 On: Tue, 02 Dec 2014 15:06:22

INSTABILITY OF LIQUID SURFACES 919

that there is a zero-order accelerated flow of the liquid layer. The first-order perturbation of this flow, which satisfies linear equations, is represented as a series or integral of nonnal modes. Some of the modes are found to be unstable in the sense that their amplitudes in­crease indefinitely with time. Among these, one or more are most unstable, in the sense that they grow most rapidly. In the case of constant acceleration the un­stable modes grow exponentially, and the most unstable modes have the largest exponents. The most unstable modes are assumed to be responsible for the breakup of the layer into pieces, which then become spherical due to surface tension. The number of pieces depends upon the number of positive and negative regions into which the most unstable mode function is decomposed by its nodal lines. Thus, it is possible to estimate the number and size of the resulting drops.

In the following section (II) the problem is fonnulated for the case of a plane layer. In Sec. III the zero-order solution is considered, and in Sec. IV the first-order perturbation is obtained. This perturbation solution is analyzed in Sec. V, where the fonnation of drops is discussed, and practical conclusions and recommenda­tions are presented. Corresponding results for cylin­drical and spherical layers will be given in another paper.

II. FORMULATION

We consider the motion of a plane slab or layer of liquid which separates two gases at different pressures. The liquid will be assumed to be incompressible and inviscid, and the motion of the gases will be neglected. The initial positions and shapes of the bounding surfaces will be prescribed, as well as the initial velocity of the liquid. The problem is to determine the sub­sequent positions and shapes of these surfaces, and the motion of the liquid.

Assuming that the motion is irrotational, the velocity u(x,y,z,t) of the liquid is derivable from a potential cp(x,y,z,t) which satisfies Laplace's equation. Thus,

(1)

(2)

The pressure p(x,y,z,t) in the liquid can be expressed in tenns of cp, the constant density p, and the acceleration of gravity g by Bernoulli's equation

(3)

It is assumed that the positive z axis points vertically upward.

The boundary surfaces are assumed to be given by

z= F,(x,y,t) i= 1,2. (4)

The subscript i= 1 corresponds to the upper surface. The gas pressure on both surfaces is assumed to be independent of x and y, and is denoted by p,(i= 1,2). Then the dynamic and kinematic boundary conditions

may be written, respectively, as

CPI- gF,_!(Vcp)2- (_I)ip-lT,2H,= p-lp, on z=F, (i= 1,2) (5)

CPz-CPxF,x-cpyFiu+F,t=O on Z=Fi (i= 1,2). (6)

In (5), T, is the surface tension and Hi the mean curva­ture of the surface i. The initial conditions will be intro­duced when they are needed.

If the flow is bounded by a cylindrical tube parallel to the z axis, the condition iJcp/iJn=O is imposed on this tube.

If the lower surface Z=F2 is a rigid surface with prescribed motion, the above formulation applies pro­vided (5) is omitted for i= 2. Therefore, this problem will also be considered, since it is so similar to the original problem.

III. ZERO-ORDER SOLUTION

If bounding surfaces are initially planes normal to the z axis, and if the initial velocity is parallel to the z axis and independent of x and y, the solution will be inde­pendent of x and y. Then, if u(t) denotes the z com­ponent of velocity, we have, from (1) and (2),

(7)

Since Fl depends on t only, we have, from (5) through (7), denoting t differentiation by a dot,

h= (u+g)Fl+!U2+p-lPi i= 1,2 (8)

i=I,2. (9)

From (9) we have, if h denotes the initial separation between the surfaces,

F2°=Flo-h.

From (8), eliminating h, we obtain, using (10),

(u+g)ph= P2- Pl.

(10)

(11)

Equation (11) yields u, and therefore u if its initial value is given, provided P2- PI is constant or a known function of t (or even of u and t). If the pressure differ­ence is constant, u is constant. However, if P2- PI depends upon F 2°, then (11) becomes a second-order equation for F2°(t). Finally b is detennined from (8), except for an inessential additive constant.

In the problem in which F 2°(t) is a prescribed moving rigid surface, u and FlO are given in tenns of FlO by (9, 10), and b by (5) with i= 1, while (5) for i= 2 is omitted, as well as its consequence (11).

IV. FIRST-ORDER PERTURBATION

Suppose that the initial boundary surfaces differ slightly from planes, and for that the initial velocity differs slightly from the constant z component assumed in the above solution. Then we may expect the sub­sequent solution to differ slightly from the above solu-

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.22.67.7 On: Tue, 02 Dec 2014 15:06:22

920 J. B. KELLER AND 1. KOLODNER

tion, which we will henceforth call the zero order or unperturbed solution. If ~ is a measure of the maximum deviations of the initial data, we assume that the solution may be written in the form

From (2) and (12) we see that cpl also satisfies Laplace's equation. Now inserting (12) into (5) and (6) we obtain for the first-order perturbation cpl and FI the equations

CPt- (ti+g)Fi+ucpz- (-1)ip-IT;'v2Fi=0 on z= F iO (i= 1,2) (13)

CPz+Fit=O on z=F;O (i= 1,2). (14)

In (13) and (14) and the sequel, the superscript one on cp and F is omitted.

To solve Laplace's equation for cp we assume that cp is a product of a function of z and t multiplied by a . function if;(x,y). We find

cp= [gl (t)ekz+ g2(t)e-kz ]if;(x,y) (15)

(Vl+k2)if;=0. (16)

If the flow is unbounded in the x,y plane, then (16) has a bounded solution for every real value of k, and the general solution for cp is an integral with respect to k of the expression on the right of (15), in which gl and g2 depend upon k.

However, if the flow is bounded by a rigid cylinder parallel to the z axis, and if this cylinder intersects the x,y plane in a curve C, then the function if; must also satisfy the condition

aif;liJn=O on C. (17)

If C is closed, then (16) and (17) have solutions only for a denumerable set of positive values of k2

, and for k2= O. In this case the general solution for cp is a series of terms of the form given by (15), summed over the various values of k.

In order to complete the solution we must find gt, g2, F I and F 2. To this end we assume

Fi= fi (t)if; (x,y) i= 1,2. (18)

Now, inserting (15) and (18) into (13) and (14) and making use of (16) we obtain

(it exp(kFIO)+Y2 exp( - kFiO) - (U+g)fi+ukgl Xexp(kFiO)-ukg2 exp( -kFl)

+(-l) ik2p-IT;fi=0 (i=1,2) (19)

kgl exp(kFP)-kg2exp(-kFl)+/i=0 (i=1,2). (20)

From (20) we find gl and g2 in terms of /1 and j2,

gl = -k-I exp( -kFIO)(jl-e-kh/2)(1-e-2kh)-1 (21)

g2=k-1 exp(kFIO) (/1-ekh j2) (1-e2kh)-l. (22)

Now using (21) (22), and (9) in (19), and combining the two Eqs. (19), we obtain

[(iJ2ft! at2) - e-kh (a2/21 at2) ]+k(u+ g) (fl- e-kh /2) + k3T IP-1 (/I+ T I-IT2e-kh h) =0 (23)

[( a2 /II at2) - ekh (iJ2 fd iJt2)]- k(u+ g) (/I-ekh/2) - k3T IP-I(/I+ T I-IT2ekh /2) = O. (24)

Equations (23) and (24) are a pair of linear homo­geneous second-order ordinary differential equations for /I and /2. The coefficients are constant only if u is constant, which case we shall treat in detail. First, however, we wish to point out the limiting forms which (23, 24) attain when h becomes infinite, namely,

(iJ2/I1 at2)+k(u+g+k2T lP-I)/I =0, (25)

(26)

Either of these is the equation for the perturbation of the surface of a semi-infinite region of liquid, in agree­ment with Taylor's result. We note that, for sufficiently large constant '11, one of the foregoing equations will have a growing exponential solution while the other will not.

If '11 is constant we seek solutions of (23) and (24) of the form

fi=Aieat i= 1,2. (27)

Inserting (27) into (23) and (24), and simplifying, leads to an equation for a and another for A 21 A I,

namely,

a4+a2k3p-I(TI+T2) cothkh - P(u+g+k2T IP-I) (u+g-k2T 2P-I) =0 (28)

A21 A I =ekh(a2+k[u+g]+k3TIP-I) X (a2+k[u+g]-k3Tw-I)-I. (29)

The solutions of (28) for a2 are given by (when T I = T 2)

ph3TI-Ia2= -Xl cothx±x(tJ2+x4 sinh-2x)1. (30)

The abbreviations used in (30) are

x=kh, i1=ph2T I-Ilu+gl. (31)

Since there are four values of a, each fi is a sum of four terms of the form in (27).

In the problem in which F 2(t) is the prescribed motion of a plane rigid surface, F 21 (t) is zero. All equations through (24) apply, but (13), (18), and (19) hold only for i= 1 and 12=0 in (21) and (22). Instead of (23) and (24) we obtain for /t the equation

(iJ2/tliJt2)+k tanhkh(u+g+k2T IP-I)fl=O. (23')

For infinite h this also has the limiting form (25).

V. INSTABILITY AND DROP FORMATION

If a2 is negative, a is pure imaginary and the corres­ponding exponential is oscillatory. This occurs when

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.22.67.7 On: Tue, 02 Dec 2014 15:06:22

INSTABILITY OF LIQUID SURFACES 921

the negative sign is chosen in (30), Only when the plus sign is chosen is there a range of x for which a2 is positive, and thus one value of a is real and positive leading to a growing exponential. In this range there is a greatest value of a corresponding to a particular value of x which we will call Xmax, since the exponential corres­ponding to this value of x grows most rapidly,

To find Xmax we first note that

(32)

Thus, if 13»(1.3)4= 1.14, (30) becomes (using the plus sign)

(33)

From (28), a2 is positive for x between zero and i3i , It has its maximum at approximately

Xmax= (i3/3)!=h[Plu+gI/3T1J!, (34)

Recalling that by (31), x=kh, we may introduce kma:x which is the value of k corresponding to Xmax, and thus to maximum instability. From (34) we have

(35)

To find the number of pieces into which the layer breaks up due to the growth of the term corresponding to kmax, we must examine the function if;(x,y) corres­ponding to kmax• It is dear that if the layer is unbounded in the x,y plane there will be many functions if;(x,y) corresponding to kmax . On the other hand, if the region is bounded, (17) applies, and there may be no solution if; corresponding exactly to km.x, although there will be many functions if; corresponding to values of k near kmax • Thus, in both cases there will be many functions if; with exactly or approximately the same value of k ( "'" kmax) , and therefore all the corresponding terms will grow at about the same (maximum) rate. Con­sequently, the exact manner of breakup will depend upon the extent to which these various terms are ex­cited by the initial perturbation.

Nevertheless, it is possible to make a rough estimate of the number of pieces into which the layer breaks. First, in the case of an unbounded region, one solution if; corresponding to kmax is

where if;= exp[2wi(xIAI +y IA2) J,

(27r IX 1)2+ (27r IX2)2 = kmax 2.

(36)

(37)

We now consider the regions of the x,y plane in which the real part of if; is positive (or negative). These are

rectangles bounded by nodal lines, and their dimensions are AJ2 and Ad2, with area A1X2/4, If Al and A2 are related by (37), the area of such a rectangle is mini­mized when X1=A2=27rY2/km a:x. Thus, the minimum area of such a rectangle is

-=--=--- (38)

It is to be expected that (38) will represent roughly the area of a piece into which the layer breaks up since all parts of the surface in this region move in the same direction. The volume of a piece will be given by (38) multiplied by the thickness h. The radius r of the sphere into which the piece will ultimately deform is then [using (l1)J

[ 91rTlh ]1 [97rT1h2 ]1 (39)

r= 2plu+g! = 21p2-PtI .

The same results are obtained if the layer is bounded by a tube of rectangular cross section, and they will also apply for tubes of almost any other shape.

On the basis of (39) three means are available for diminishing drop size, namely,

(a) Reducing the layer thickness h; (b) Reducing the surface tension T1 by using wetting

agents; (c) Increasing the explosive pressure P2.

Of these the first two seem reasonable, while the last appears impractical. We note from (39) that diminishing h has a twofold effect: the acceleration is greater and the pieces are thinner.

To estimate the time required for breakup to occur, we assume that the surface perturbation, given essen­tially by (27), reaches the value h. This gives for the break-up time

(40)

In (40) A is the initial perturbation of the surface, amax was obtained from (33) and (34), and the last equality follows from (11).

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.22.67.7 On: Tue, 02 Dec 2014 15:06:22