This document presents a linear stability analysis of the Rayleigh-Taylor instability in elastic liquids. It considers the stability of an incompressible elastic fluid at rest above a free surface. First, it uses the Maxwell model and neglects inertial effects, finding a characteristic equation with a critical wave number. However, it finds an unbounded growth rate at a certain Weissenberg number. It then shows that introducing inertial terms or changing to the Jeffreys model smooths out this singularity. Asymptotic expansions are developed to show how the singularity is approached in both cases.
This document presents a linear stability analysis of the Rayleigh-Taylor instability in elastic liquids. It considers the stability of an incompressible elastic fluid at rest above a free surface. First, it uses the Maxwell model and neglects inertial effects, finding a characteristic equation with a critical wave number. However, it finds an unbounded growth rate at a certain Weissenberg number. It then shows that introducing inertial terms or changing to the Jeffreys model smooths out this singularity. Asymptotic expansions are developed to show how the singularity is approached in both cases.
This document presents a linear stability analysis of the Rayleigh-Taylor instability in elastic liquids. It considers the stability of an incompressible elastic fluid at rest above a free surface. First, it uses the Maxwell model and neglects inertial effects, finding a characteristic equation with a critical wave number. However, it finds an unbounded growth rate at a certain Weissenberg number. It then shows that introducing inertial terms or changing to the Jeffreys model smooths out this singularity. Asymptotic expansions are developed to show how the singularity is approached in both cases.
Journal of Non-Newtonian Fluid Mechanics, 49 (1993) 13-22
Elsevier Science Publishers B.V.. Amsterdam
13 Rayleigh-Taylor instability in elastic liquids L.S. Aitken and S.D.R. Wilson * Department of Mathematics, University of Manchester, Manchester Ml3 9PL (UK) (Received October 28, 1992) Abstract In this paper the problem of the stability of an incompressible elastic fluid at rest above a free surface is considered by carrying out a linear stability analysis. In the first case considered the Maxwell model is used to include elastic effects but inertial effects are neglected. A characteristic equation for the growth rate is found showing a dimensionless critical wave number. However, for each fixed wave number in the region of instability, this model is found to predict an unbounded growth rate at a certain value of the Weissenberg number. It is then shown how the introduction of inertial terms smooths out this singularity as does the change to a model which introduces a second elastic parameter, the linear Jeffreys model. Asymptotic expansions are developed to show how the singularity is approached in both cases. Keywords: elastic liquids; inertial effects, linear stability analysis; Maxwell model; Rayleigh- Taylor instability 1. Introduction The instability of the interface between two superposed Newtonian fluids of different densities at rest was initially studied by Rayleigh [ 11. Taylor [2] included the effect of a constant acceleration acting perpendicularly to the interface and concluded that if the acceleration is directed from the less dense to the more dense medium then any slight disturbance to the interface will grow exponentially with time. Subsequently, Bellman and Pennington [3] examined the effects of surface tension and viscosity on this instability and discovered the presence of a critical wave number with the property that the interface is stable or unstable depending on whether the wave number is greater than or less than this critical wave number. The special * Corresponding author. 0377-0257/93/$06.00 0 1993 ~ Elsevier Science Publishers B.V. All rights reserved 14 L.S. Aitken and S.D.R. Wilson 1 J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 case when the density of the lower fluid is negligible is equivalent to a fluid at rest above a free surface and is referred to as the Rayleigh-Taylor instability. Where non-Newtonian fluids are concerned very little work has been done on this problem, even in the linear case. Saasen and Tyvand [4] derived a characteristic equation for the rate at which the system departs from equilibrium but unfortunately were mistaken in their analysis of the normal stress condition. In this paper we aim to rectify this mistake and by a better choice of dimensionless variables produce a recognisable character- istic equation which shows critical wave numbers and the effect of the inclusion of elastic parameters. We include surface tension, viscosity and fluid depth in our calculations, restrict ourselves to incompressible media and use only the linearized equations of hydrodynamics. 2. Statement of problem The basic problem of a fluid at rest above a free surface is essentially two dimensional. Coordinates Oxy are chosen with Ox horizontal and Oy verti- cally downwards, with the undisturbed free surface at y = 0 and a rigid boundary at y = -h. From the general equations of motion and a suitable rheological model we carry out a standard linear stability analysis and find a characteristic equation for the growth or decay rate of the disturbance in terms of the wave number k and other parameters in the problem. We first examine the characteristic equation for the case in which inertial effects are negligible and the constitutive equation is that of the Maxwell liquid. As expected, a critical wave number k, is shown to exist, such that waves are stable or unstable depending on whether k > k, or k < k,. However, for fixed wave numbers in the region of instability k < kc, there is a critical Weissenberg number IV,, at which the growth rate is infinite. We then show that this singularity is removed when either the constitutive equation is generalized to that of the Jeffreys model or inertial effects are introduced. To complete our analysis we develop asymptotic expansions which show how the singularity is approached in both these cases. 3. Mathematical formulation In the underlying state, whose stability is being investigated, the fluid is at rest and there is a hydrostatic pressure field. In the following sections we present the general form of the equations of motion and boundary conditions, after linearization about this underlying state in the usual way. These are then converted into dimensionless form before solutions are developed. L.S. Aitken and S.D.R. Wilson 1 J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 15 3.1 Equations after linearization These take the forms: equation of momentum i3U p,, = -gradp +divr, equation of continuity div u = 0, constitutive equation r++2p e+i,;. ( > (1) (2) where u is the (two-dimensional) disturbance velocity vector. We denote the x and y components of this vector by u, u respectively. Also z is the extra-stress tensor, e is the rate-of-strain tensor, p and ,D are the density and viscosity of the fluid, and p is the pressure due to the disturbance, i.e. the difference from hydrostatic pressure. Finally ,I1 and A2 are the usual relax- ation and retardation times respectively. Equation (3) is the Jeffreys model. The linear model has been shown to be sufficient in our case as we are studying a perturbation of a state of rest; see Joseph et al. 151. Since the Jeffreys model incorporates both elastic and viscous responses we can expect that for our problem this model is suffi- ciently general to capture all the typical behaviour of a general linear model. Accordingly, our results will not be seriously model dependent. 3.2 Boundary conditions after linearization These take the forms: no-slip condition u=v=O on y=-h, kinematic boundary condition aY u=~ on y=O, continuity of tangential stress z _uy = 0 on y = 0, continuity of normal stress p=-g?l ax2 -Pm + zyv on y = 0, (4) (5) (6) (7) 16 L.S. Aitken and S.D.R. Wilson /J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 where T is the surface tension, g is the acceleration due to gravity and the fluid surface has displacement q = ~(x, t). Note that in eqn. (7) it is necessary to estimate the total pressure (that is, the background hydrostatic pressure plus the disturbance p) at the disturbed interface y = q; on linearization about y = 0 this produces two terms, the term p on the left and the term -pgq on the right. It is in the normal stress condition that Saasen and Tyvand [4] appear to be in error. They chose to adapt an earlier analysis of the stable case and omitted a change of sign (of the first term on the right of eqn. (7)). 3.3 Dimensionless variables We choose a system of dimensionless variables based on the length scale a = (TM . j2 This represents a balance between the stabilizing effect of surface tension and the destabilizing effect of gravity and is chosen so that the dimensionless solution will reflect this balance. A dimensionless time, velocity, pressure and extra-stress tensor are intro- duced, defined by * t=!l! pga u =u**, p =p*pgb, z=+pgb. P where the asterisk denotes the dimensionless quantity and b is the arbitrary disturbance amplitude. These are substituted into the basic equations and boundary conditions. Omitting the asterisks from now on we obtain Rg = -gradp +divr, div u = 0, (9) where R =ga3p2/p2, W, = Alpga/p, W, = I,pga/p are the dimensionless Reynolds number and Weissenberg numbers, with boundary conditions u=v=O on u=--H, afl =dt on y=O, z y = 0 on y = 0, (11) a2r -P-V++_Vy=s on y=O, where H = h/a. L.S. Aitken and S.D.R. Wilson /J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 17 3.4 Solution procedure As plane surface waves are being cbnsidered we assume, in the usual way, that the dependent variables take the form of an amplitude factor multiplied by exp(ikx + cot), where k is the wave number and w is the growth rate. After substitution the constitutive equation (10) becomes L, z = 2L2e, (12) where L, = 1 +wW, and L2= 1 +wW,. After similar substitutions we eliminate terms to obtain an ordinary differential equation for v(y) : D2-,7&? > v =o, 2 (13) with boundary conditions v=O,Dv=O on y=-H, (D*+k*)v=O on y=O, RwL, D2-r- 3k2 Dv-k2 2 > (k2 - 1) L L,v=O on y=O, 2 where D denotes d/dy. 4. Results and discussion 4.1 R=O, W2=0, W,#O In this case we assume negligible inertial effects and that the constitutive equation is that of the Maxwell liquid. The equations and boundary conditions simplify somewhat and after lengthy but straightforward calculations we find that, for a non-trivial solution to exist, 0 must satisfy ( 1 - k)(sinh kH cash kH - kH) w = 2k(k2H2 + cosh2 kH) - W, ( 1 - k2)( sinh kH cash kH - kH) (14) We note the presence of the dimensionless critical wave number k, = 1. For values of k > k,, cc) is negative and thus is a decay rate implying stability. This holds for all values of the Weissenberg number and is due to the fact that for small wavelengths, curvature is large and surface tension effects dominate the destabilizing gravitational effects. Note that this critical wave number is the same as in the Newtonian case, i.e. the stability boundary is unchanged by the presence of elasticity. 18 L.S. Aitken and S.D.R. Wilson 1 J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 For k <kc we expect w to be positive and the system to be unstable but when we examine the characteristic equation we note that there is a singularity in the solution. For each fixed wave number this singularity occurs at a particular value of the Weissenberg number WC given by 2k(k2H2 + cosh2 kH) K = ( 1 - k2)(sinh kH cash kH - kH) (1% As IV, increases from zero, through IV,,, the growth rate tends to infinity and then jumps to large negative values. This behaviour is unphysical and is undoubtedly an artificial effect of overidealization of the problem. Similar effects were noted by Wilson [6] in the Taylor-Saffman fingering instability. The root of the problem can be traced to the well-known fact that the Maxwell fluid, in the absence of inertial effects, has a purely elastic instantaneous response and exhibits a finite jump in strain in response to a sudden change in stress. The stability problem considered here can be thought of as an initial value problem in which the short-term response to a small, but sudden, change in configuration is calculated and might, therefore, be expected to show analogous behaviour. The instantaneous jump in strain mentioned above is smoothed out by the inclusion of a finite retardation time, as in the Jeffreys model, which forces a purely Newtonian element in the short-term response. It will also clearly be smoothed out by inertial effects. In the next two sections we introduce these effects, one at a time, in order to examine in detail how the singularity is modified. 4.2 R=O, W,#O, W, #O In this case we assume that inertial effects remain negligible but examine how the introduction of a second elastic parameter, using the Jeffreys model, removes the singularity. After a straightforward calculation we find that w must satisfy 2k(k2H + cosh2 kH) FV, co2 + [ 2k(k2H2 + cosh2 kH) - W, ( 1 - k)( sinh kH cash kH - kH)]co - (1 - k2)(sinh kH cash kH - kH) = 0. (16) This can be written as Fv302+ 1-s 0 -f=O, ( > C C (17) where IV, is the critical value of IV, given by ( 15). Note that when IV2 = 0 we recover (14). L.S. Aitken and S.D.R. Wilson 1 J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 19 Y Fig. 1. The positive root of ( 17) as a function of W, , for various fixed values of W,. The value of WC has been set arbitrarily at WC = 1. When k < 1 this quadratic has two roots of opposite sign. When IV, + 0 the positive root goes to infinity for values IV, > IV, (and the negative root goes to infinity for IV, < IV,). This explains how the singularity of the previous section can be unfolded by the parameter II,. Figure 1 may help to clarify the structure. This shows a graph of the positive root of (17), as a function of IV,, for various values of I%,; the value of IV, has been set arbitrarily at IV, = 1. The singularity is removed when IV, > 0, but we see that co + cc as IV, + 0 with IV, fixed and > IV,. The asymptotic form of this limit is easily obtained from (17): ( > _-1 1, W 1 cc)= WC w, (W,- WC) (18) for WI > WC. When W, = WC this expansion obviously fails and we have instead 1 ( > iI2 co= w,w, . (19) Note that the approach to infinity, as W, + 0, is slower in (19) than in (18). 4.3 W,=O, R#O, W,#O In this case inertia terms are introduced but the retardation parameter 1, = 0, i.e. the Maxwell model is retained. To simplify the algebra only the 20 L.S. Aitken and S.D.R. Wilson f J. Non-New~o~~~n Fluid Me&. 49 (1993) 13-22 case of infinite depth is considered. Note that this limit (liT -+ CO) clashes with the limit k -+ 0 so it will not be possible to derive long-wave approxi- mations from the formulas given below. Equation ( 13) when modified has the general solution v ;: A ek- + B eeky + C er,v + D e-y, (20) where r2=k2+RCO{1 4wlV,), (21) and we choose the root for Y with positive real part. Further calculation gives the characteristic equation for CO: 4k4+R202(1 +cQW*)~+4k2RCO(l +ct,JV1) -NC{1 -P)(l +COIV~)2=4k3Y_ (22) Note that for k = 1, w = 0 is a solution, so the stability boundary remains unaltered by the introduction of inertia terms. Equation (22) can be solved n~e~cally if required. Here we wish to examine o for small R and so look for an asymptotic form for co using the straightforward expansion for R -+ 0 u>eoo+Rcol+.--. We find that (1 -k2) (1 -k2)2 ~~2k-W,(l-k~)+192k[2k-W,(l-k~)]3+~~. From this expansion it can be seen that the critical wave number is kc =1, as before, with instability for k <kc. Again in the region of i~~stability k < 1 there is a critical value of PV,, WC, given by 2k WC=:------- 1 -k2 cm at which CO becomes infinite. This is of course the limit of (15) as kH -+co. This expansion is correct for Ity, < r;V,; however, it fails as IV, approaches w, from below so a different expansion is required for the region ?V, 2 fVC. For W, > WC we find, after some investigation, that where coo is a solution of CO; r/r/: + co;[4k2W, - (I - k2)kWf] +4k4 =4k3(k2 +u$ Wl)2. (27) L.S. Aitken and S.D.R. Wilson /J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 21 Fig. 2. The positive real root of (29) as a function of W, / WC. It can be shown by a graphical argument that this equation does indeed have a positive real root when IV, > IV,. In order to obtain this root it is convenient to rewrite it using the substitution CD; = k=e/ W,) (28) and it is found that 8 satisfies the equation 2% e*+ 4-w ( > e + 4 = 4( 1 + e) i/2. C The required positive root I!$, can be found easily by numerical methods; it is given as a function of W, / WC in Fig. 2. For R + 0, in the region W, > WC, co has the approximate form (30) This again fails at W, = WC. To investigate the behaviour of the relevant root of (22) in the limit R + 0, W, = WC we first recall ( 18) and ( 19) from which we may conjecture that the limit co + co will be approached more slowly than R _ I /2 as shown in (30). If so, then we have oR and coR both small as R + 0 and the appropriate expansion of (21) can be substituted into (22). After some calculation we find (31) 22 LX Aitken and S.D.R. Wilson / J. Non-Newtonian Fluid Mech. 49 (1993) 13-22 5. Concluding remarks The characteristic equation for the growth rate was found for the case where the constitutive equation was that of the Maxwell model and inertial effects were considered negligible. The dimensionless critical wave number, k, = 1, was found to be the same as in the case of a Newtonian fluid. However, in the region of instability, k <k,, singularities were found in the solution for certain values of the Weissenberg number WC. Further analysis showed that when inertia terms or a second elastic parameter were introduced (R # 0 or IV, # 0) this singularity was removed. In each case asymptotic expansions were developed for the regions IV, < IV,, IV, = IV,, IV, > IV,, and these showed how the singularity was smoothed out for small values of R or W2. It is well known that elastic liquids with similar behaviour in steady shear may behave very differently from each other in unsteady flows. Also, the anomalous behaviour predicted by Maxwell models has been discussed by Joseph and Saut [7]. The present calculations have shown that a further anomaly exists, namely the prediction of unbounded growth rates. It would be interesting to consider a possible experimental comparison in this context of the stability of two liquids, one Newtonian and the other highly elastic, with similar values of the steady-shear viscosity. References 1 Lord Rayleigh, Investigation of the character of an incompressible heavy fluid of variable density, Proc. London Math. Sot., 14 (1883) 170-177. 2 G.I. Taylor, The instability of liquid surfaces when accelerated in a direction perpendicular to their planes, Proc. R. Sot. London Ser. A, 201 (1950) 192-196. 3 R. Bellman and R.H. Pennington, Effects of surface tension and viscosity on Taylor instability, Q. J. Appl. Math., 12 ( 1954) 151- 162. 4 A. Saasen and P.A. Tyvand, Rayleigh-Taylor instability and Rayleigh type waves on a Maxwell fluid, J. Appl. Math. Phys. ZAMP, 41 (1990) 2844293. 5 D.D. Joseph, A. Narain and 0. Riccius, Shear-wave speeds and elastic moduli for different liquids, Part 1. Theory, J. Fluid Mech., 171 ( 1986) 289-308. 6 S.D.R. Wilson, The Taylor-Saffman problem for a non-Newtonian liquid, J. Fluid Mech., 220 (1990) 4133425. 7 D.D. Joseph and J.C. Saut, Change of type and loss of evolution in the flow of viscoelastic fluids, J. Non-Newtonian Fluid Mech., 19 (1986) 237-249.