LaIs.
y. Scheepsbouwkundé
Tech&sche Hogeschool
353
Delit
On Kelvin-Helmholtz instabifity in a rotating fluid
By HERBERT E. HUPPERT
Institute of Geophysics and Planetary Physics,
University of California, La Jolla
(Received 16 August, 1967)
Chandrasekhar's (1961) solution to the eigenvalue equation arising from the
KelvinHelmholtz stabifity problem for a rotating fluid is shown to be incorrect. The unstable modes are correctly enumerated with the aid of Cauchy's principle of the argument. Various previously published solutions using Chandrasekliar's analysis are corrected and extended.
Introdüction
Chandrasekhar (1961, § 105) devotesa section of his treatise to the investiga-tion of the effect of rotainvestiga-tion on the development of KelvinHelmholtzinstability. Two uniform, horizontally-superposed fluids of different densities are in relative horizontal motion and are rotating with uniform angular velocity about a vertical axis. Perturbing the solution with a travelling wave in the direction of streaming,
imposing the boundary.conditions at the interface, and linearizing,
Chandrase-khar obtains the eigenvalue equation for the wave speed. He then uses a graphical
method, together with the fundamental theorem of algebra, to enumerate the eigenvalues. His argument necessitates determining the singular points of the eigenvalue equation. We show that Chandrasekhar's determination of these
/
singular points is incomplete, leading toerroneous results. With the aid of-y
Cauchy's principle of the argument, we enumerate the eigenvalues and presenta simple, sufficient condition for stability.
Alterman, in a series of papers (1961 a, b, c) dealing with the
KelvinHelmholtz
stability problem under various force fields, uses Chandrasekhar's results and
generalizations of his method, in consequence of which her results are incorrect.
We present the correct results and, inone instance, extend our analysis to solve
the problem for a more general flow configuration than thattreated by Alterman.
Chandrasekhar's analysis
Chandrasekhar considers two uniform, superposed fluids of densities Pi' P2 having velocities U1, U2 in the x direction and ina state of uniform rotation about the z-axis with an angular velocity £1 [In what follows a subscript 1 (2) refers
to the lower (upper) fluid; Pi is greater than P2.] Imposing upon the steady-state solution a small disturbance whose dependenceon z and t is given by
exp [ik(xct)],
where k is the wave number and c the wave speed, linearizing the equations of
23 FluidMech.33
ARCHIEF
J. Fluid Mech. (1968), vol. 33,part 2, pp. 353-359
354
H.E.Huppert
motion, and applying the conditions of continuity of pressure and normal
velocity at the undisturbed interface, Chandrasekhar obtains the
eigenvalueequation
- 6)2 [1- 4l2k-2(U - c)] + a2 (U2 -
c)2 [1- 42k-2(U2 -
C)_2J1-g1c-'(a-a2)-kT(p1+p2)' = 0,
(2.1)= pJ(pi+p)
(i = 1,2), (2.2a,b)where
roots of (2.1) must be taken to have positive real parts since
the product of wave number and the inverse ofthese real parts represents the rate of decay
g is the acceleration due to gravity, and T is the surface tension. The squareof the disturbance with increasing distance from the interface.
Applying the transformation
-c = g(g/k)+, 172-c = i(g/k)f, (2.3a,b)
originally due to Taylor (1931), and neglecting surface tension,Chandrasekhar reduces the problem to the simultaneous equations
(2.4)
= a1a2
g -
= V, (2.5) and (2 = 4Q2/(gk), V = (U1- U2) (k/g). (2.6 a, b) whereIn order to apply the fundamental theorm of algebra, which is not directly applic-able to (2.4), Chandrasekhar introduces the following set of four equations:
- w2g2)t + cc22(1 - w2_2)l = a1 - a2, (2.7)
w2g-2)1 - a2ij2(1 - w_2)t = a1 - a2, (2.8) g2(1 _w2g-2)1 + a2172(1 _w2ìj_2)t = a1-a2, (2.9)
-a1(1 _w2g2)_a2n2(1
_w_2)l a1 a2.
(2.10)He then states that, from the fundamental theoremof algebra, equations (2.7) to (2.10) 'have a total of exactly eight roots ...[which] are continuous functions of
the parameters of the equation[s] except at the singular points (± g0, ± w) and
(± w, ± ), ... where and i are determined by the equations
a1g2(1 _w2g-2)t = a1-a2 and a22(1
_w2_2)t = a1-a2'.
(2.11)Using these two facts, Chandrasekhar determines the number of roots of each of (2.7)-(2.10) as follows.
The equations are solved for the particular case
V = O ( = ).
jVis then
increased until the first singular point is reached, IVI =
g0-w
V1, say. At thispoint the equations may exchnnge roots Using the enunciated theorem and the
graphs of the equations, Chandrasekhar determines thenumber of roots of each
equation at IVI = T7+. This procedure is then continued, Chandrasekhar
con-sidering the exchange of roots at each of the singular points quoted above.
However, (±
,0) and (0, ± ) are also singular points,
in consequence ofwhich Chnndrasekhar's results are incorrect.tFurthermore, the inclusion of these
Prof. Chandrasekhar informs me that he is aware of this error and will present an
355 singular points renders the method inapplicable, as it is not possible, by Chan-drasekhar's method, to determine theexchange of Ñots between (2.7) and(2.8)
at (±
,O) and between (2.7) and (2.9) at (O,± 21o).In the next section
we use Cauchy's principle of the argument to yield thecorrect enumeration of the eigenvalues of (2.1).
3 Cauchy's principle of the
argument
Introducing the transformations
c=(Uj+U2)+V, V=(U1U2),
(3.la,b)
= oct,
ic = kV2/g, z = 2Q/(kV), = k2T/(p1g), (3.2a,b,c,d)we write the eigenvalue equation as
E(a) 0T 1 +ox(a+ 1)2[1 z2(a+ 1)_2]+K(8-
1)2 [1 z2(a 1)2j1
= 0, (3.3)where the square roots are to have positive real parts. (From here on weassume
V, and hence z, to be positive; the results are not altered if V is negative.)
'w
{s}
FIGuRE 1. The contour in the 8-plane to whichCauchy's principle of the argument
is applied. D may be to the right of E and the branch cuts overlap.
To render E(a) analytic, we introduce branch cuts in the 8-plane extending along the real axis from
- 1x to 1+ z and from 1x to 1+ x.
In order to determine the solutions of (3.3) we apply Cauchy's principle of the
argument: for any function f(z), analytic within and on a closed contour ', the number of zeros minus the number ofpoles of ¡(z) within ' is (1/2ir) times the
increase of argf(z) as z traverses once in an auticlockwise direction (see Copson 1962, § 6.2). We apply this principle to F(s) and take Sf to be the real axis of the
8-plane, indented above the branch cuts, anda semi-circle whose radius tends to
infinity in the upper halfplane. A typical contour is shown in figure 1.
To construct the map of ' onto the F-plane (the OauchyNyquist diagram),
we define x(o-) as that value of x for which
= o-[2(s+ 1)2x2]{(8+1) [lx2(s+ 1)-2]*}_1
±[2(s l)2x2]{(s_ 1) [1x2(& 1)_2]}1
= Ø (34)356
H.E.Huppert
has a double root in
i ±x <8 < i-x (or, equivalently, that value
of z forwhich F"(s) = O in this range of a). x1(o-), which isdetermined numerically, is
shown in figure 2. For sin the range (-1+x,1-x),
F(s) has one maximum ifo.
FIGm 2. The curves x1(o), x2(o).
z x1(o) and two maxima and oneminimum if z <x1(o). We label the valuesof
s, at the maxima a
and b, E(a)
E(b)F, and at the minimum m,
F(m)We also define x2(ci), determinednumerically and shown in figure 2, by
for
x=x2(o).
It can be shown that z
x2(o) implies E F1. In addition, the stipulation thatthe square roots of (3.3) havepositive real parts requires that forall z
[1_x2(s+1)_2Il
j[x2(s1)_2_i]lsgn(8+i)
(son BD), (3.5)[i x2(s 1)_2j = ix2(s
1)2_ 1]sgn(s 1)
(s on EG). (3.6)The construction of the Cauchy-Nyqiiist diagrams, examples ofwhich are shown in figure 3, is now straightforward.
To enumerate thé complex eigenvalues, we determine the number oftimes the Cauchy-Nyquist diagram encircles the origin [from Cauchy's principle and the absence of poles of F(s) within , this is the number of pairs of complex
conjugate eigenvalues]. The positiòn of the origin in figure 3 isdependent upon the particular values of X, o, K, r. Exploring all possibilities (there are
approxi-mately thirty), we determine the number of complex eigenvalues as shown in table i [F_1
F( - 1) therein]. From the table, we see that: there cannot be
(e)
{F
(f)
FIOuRE 3. The CauchyNyquist diagrams. (a) x <z1, X2. (b) z2 <z <z1. (e) z1 <x< z2.
(d) z1,z2 <z < 1. (e) 1.< z < 2. (f) z> 2.
(a) (b)
(c) (d)
358
H.E.Huppert
instability for x 2; a sufficiönt condition for stabffity in z < 2 is
TABLE 1. The distribution of complex éigenvalues. N is, the number of pairs of complex
conjugate eigenvalues in the indicated range of n
For max (z1, x2)
z
2 this is also a necessary condition. For z < max (z1,z2), we see from lines (iii)and (iv) of table 1 that there may be a region of stabifity for which (3 7)is not satisfied Returning to physicalvariables, we find that asuffi-cient condition for stabifity is.
(U1 - U2)2 2Q2k-2 + k-2{44 + k2pj2[o1 - P2) +k2T]2}. (3.8)
A table enumerating the real eigenvalues can
be easily obtained, but is of
excessive length and hence is omitted. It can be simply seen from the Cauchy-Nyquist diagrams, however, that the numberofreal eigenvalues varies between zero and four. We eau also show that the total number of eigenvalues is nevermore than four Finally, we
note from the Cauchy-Nyquist diagramsthat if
<o
1?(1 +x) (3 3) has no solution whatever, in which case the original linearized perturbation equation does not admit adiscrete spectrum solution.Generalizing Chandrasekhar's argumeat to include the effect of surface
ten-sion, Alterman (1961 a) asserts thatthe system is unstable to long waves. From (3.8) we see that the' actual conditionfor stability is
(U1_U2)2<42k-2+O(1)
(k-i-0), (3.9)and hence there is stability to a long wave length disturbance.
In a later publication, Alterman (1961 b) considers the problem of two hetero-geneous fluids, with horizontal velocities U1, U2inthe same direction, separated by a horizontal interface at z = 0,the densities being given by
p
=p1e_ßZ (z < 0), p = P2e (z> 0). (3.10 a, b) Invoking the Boussinesq approximatiôn, she shows that the eigenvalue equation governing stabffity is' formally equivalent to (2.1) if 2 is replaced by ßg. The correct sufficient condition for stability is hence given by (3.8) once thisreplace-By stability here we mean stability to an exponentially growing disturbance; the
as-sumed form of the disturbance riles out any possibility of investigating algebraic instability,
for which an initial value approach is required.
In Chandìasekhar's notation, this becomes <0after setting r = 0.
1-cT+'r ? 4,C[1_(x2/4)]
(z <2).
F
N (i)[2, )
.. O (ii) (0,2) F_1 <0 0 (iii) (0,x2) Fa O,Fi < O O (iv) (0,x1) Fb>_0,Fm<O O (y) [1,2) F1>O 2 (vi) (x2,1)Fj>Q,Fa<0
2 (vii) Otherwise i359
mént has been made. Generalizing Alterman's model such that the fluid densities
are given by
p = p1e-ftic (z < O), p = p2eft2t (z> O), (3.11a,b)
we obtain the eigenvalue equation
- e)2 [1 -fi1gk-2(U1 - c)-2]t + 2(U2 - e)2 [1 -fi2gk-2(U2
--glc'(c--c2)-kT(p1+p2)-1
= 0, (3.12) and the eigenvaJues can be enumerated in the same manner as before. For the sake of brevity it suffices to saythat (3.8)is a sufficient condition for stabifity ifis replaced by (g/4) min (fi1, fi2).
In anotherpaper, Alterman (1961 c) obtainsthe eigenvalue equation pertinent
to the firnd system ongmally considered by
(Jhandrasekhar, with the added condition that the fluid be a perfect conductorunder the influence of a uniform, horizontal magnetic field of intensity H. In the limit as the wave lenSth of the disturbance tends to iuithiity, Alterman obtains the equation (3.12) withfig
replaced by 4û2 + (pH21c2/2n'p) (i = 1,2). Her sufficient condition for stabifity should be replaced by
(U1 - 172)2 g2-2 + (pR2/47rp1)+ k-2{4[L2 + (uHSkS/87rp1)]2 + k2pj2 {g(p1
- P2) +k2TJ2}i (3.13)
4. Conclusion
We conclude that an eigenvalue equation of the form (2.1) arises in various Kelvin-Helmholtz stability problems and that applying Cauchy's principle of
the argument is a simple and efficient method of enumerating the eigenvaiues.
The author wishes to thank Dr J. W. Miles for his oft-given advice and the University of Sydney for the award of a University of Sydney Post-Graduate
Travelling Fellowship. This research was sponsored in part by National Science
Foundation grant GP-2414.
REFERENCES
AVrsuÇAN, Z 1961 a Effect of surface tension on the Kelvin-Hehnholtz instábility of
two rotating fluids. Proc. NaU Acad. Sci. U.S.47, 224-7.
AL'rFrRMAN, Z. 1961 b Kelvin-Ee]mholtz instábility in media of variable density. Phy8.
Fluids, 4, 1177-9.
Azw, Z. 1961 c Effect of magnetic fieldand rotatiòn on Kelvin-Héhujholtz insta. bility. Phys.. Fluids, 4, 1207-10.
CHA1WBASEXLuAB, S. 1961 Hydrodynamic and Hydromagnetic Stability. Oxford: Claren.
don Press.
COPSON, E. T. 1962 The Theory of Functions of a Complex Variable. Oxford: Clarendon
Press.
TALOn, G. I. 1931 Effect of variation in density on the stability of superposed streams