A numerical elliptic plates

scheme

for unsteady

flow of a viscous

fluid between

A. Aziz (1) a n d T. Y. Na (2)

ABSTRACT

A new continuation method has been developed to solve the nonlinear eigenvalue problem describing the unsteady, squeezing flow of a viscous fluid between elliptic plates. Unlike the numerical schemes previously used (e.g. homotopy algorithm), the present scheme is conceptually simple, noniterative, insensitive to the first approximation and works for all values of squeeze number S characterizing the flow. The numerical results compare extremely well with those obtained with sophisticated schemes. Since existing numerical data are limited to three sparsely spaced values of S, additional data are reported for systematically spaced values of squeeze number S and ellipticity parameter ft. Although the scheme has been applied to a specific problem, it appears potentially capable of handling a variety of nonlinear eigenvalue problems.
In the present paper, we examine the case of elliptic plates and solve it by a new continuation method. The scheme has the attractive feature of being noniterative and gives accurate solutions for all values of S. Since Wang and Watson [3] present limited numerical data (because of high computational cost of their scheme) we present data for systematically spaced values of S (in the range 0 to 25) and fl (in the range 0 to 1). The computational effidency of the scheme can be judged from the fact that all the data presented here were generated within 15 seconds of CPU time on Amdahl 470V/7 system.

1. INTRODUCTION The predictk;n of flow field when a viscous fluid is squeezed between parallel plates is of fundamental importance in the study of unsteady loading of bearings and in viscometry. The flow is usually characterised by squeeze number S, which indicates the relative importance of inertial and viscous forces. At low squeeze rates (S ,~ 1), the inertial effects can be ignored and the classical Reynolds' equation can be used [1]. However, at high squeeze rates (S ~ 1), the inertial terms dominate the viscous term and one must solve the unsteady Navier-Stokes equations to describe the flow satisfactorily. The only feasible approach is to use a numerical scheme as typified by the work of Grimm [2]. Considering the flow between elliptic plates, Wang and Watson [3] recently found that, ff the normal velocity is specified as proportional to (1 - at) -1/2, where t is time and ct is a constant, then a similarity solution is possible. The analysis leads to a difficult eigenvalue problem involving two coupled third-order, nonlinear boundary value problems. The equations contain two parameters : squeeze number S and elliptidty parameter ft. To obtain a numerical solution, Wang and Watson introduced a new homotopy algorithm, which is globally convergent and does not necessitate a good initial approximation to ensure convergence. Despite this, the scheme failed to converge beyond S = 20, and a standard imbedding technique had to be employed to generate solutions for S> 20.

2. PROBLEM

For squeezing flow of a viscous fluid between elliptic plates, with normal velo city pro Portional to (1 -/tt )-1/2, the unsteady Navier-Stokes equations can be reduced to the similarity form as [3]

f" + k = s[2f" + nf" +

(f,)2 _

f. (f + g)] (1)

g ,, + flk = S[2g" + ~g" + ~1- (g92- ~1- g , , ( f + g)] (2)

fro) = g(O)= f"(o) = g" (0) = 0

(3)
(4)

where the symbols are as defined in [3] and neednot be repeated. The primes denote differentiation with respect to /.

(1) A. Aziz, Department of Mechanical Engineering, College of Engineering, P. O. Box 800, Riyadh, Saudi Arabia.
(2) T. Y. Na, D e p a r t m e n t o f Mechanical Engineering, University o f Michigan-Dearborn, Dearb o r n , Michigan 48128, USA. Journal of Computational and Applied Mathematics, volume 7, no. 2,1981. 115

3. SOLUTION METHOD Since the eigenvalue k is a constant in equations (1, 2) we introduce an additional differential equation for k as in [4],
k'= o

and n (primes) stands for nth derivative with respect to '0. The solution scheme now proceeds as follows : (1) A first approximation as given by equation (6) is assumed. (2) The functions f, f', f" on the left hand side of equation (10) are replaced by f0' f0 and f~' respectively. Similarly, the functions g, g" and g" on the lefthand side of equation (11) are replaced by go, g~ and

(5)
(6)

Let us assume the firstapproximation as f0(rl)= g0(n)= sin ( 2 n), k 0 = I

where f0 and go have been chosen as they satisfy the boundary conditions (3, 4). Indeed, any functions that satisfy the boundary conditions can be chosen as a first approximation and the method would eventually converge to the same solution. The value of k 0 = 1 is arbitrary. Next we introduce a variable r and write equations (1, 2, 5) as 1 (f.)2 _ ! 2 f " ( f + g)] f " + k - S[2f' + ~f" + -~-

6".
(3) It is noted that equations (10-14) now constitute linear boundary value problems in F and G and can be solved noniteratively by the method of superposition [5] as follows. Since there are three missing initial conditions, we introduce three constants a, b, c and write the solutions as F=F 0+aF l+bF 2+ cF 3

(16)
(17)

{fo"+ ko-SI2fo

+ nfo +:1 (f~)2 (7)

G = G 0 + a G l + b G 2+ cG 3 K=K 0+aK l+bK 2+ cK 3

(18)

-:f;(f0 + So)l}
1 (g5 2 _ l g , , ( f + g)] g ,- + 3k - S[2g" + ng" + :

= (1 -~) {g: + ~k 0_ s[2s6 + ngo + :

1(~)2 (8)
(9)

Substituting equations (16, 17) into equations (10, 11), a total of eight equations for F0,F1,F2,F 3 and G0~ G 1, G 2, G 3 can be readily obtained. Further, ffwe set the missing initial conditions as
F'(0)=a, G'C0)=b, g(0)=c it follows from the bounclary condition (13)
F0(0 ) = 0 FI(0 ) = 0 F2(0 ) = 0 F:(0) = 0

(19)

F3(0) = 0 (20)
F~(0) = 0 (21)
F~(0) = 0 (22)

F6(0)= o
F{~(0) = 0
G0(0 ) = 0

Fi(0)= 1
F]~(0)= 0
G1(0) = 0

It is reasonable to expect that if r changes from 0 to 1 the functions f, g and k will change from their first approximations f0' go and k 0 to their true values. To establish the dependence of the solution on r, we differentiate equations (7, 8, 9) and boundary conditions with respect to r. The result is
F " + K - S[2F'+ rlF" + f'F'-IF"(f+g)-~f"(F+G)] 2 ~.

F~'(0) = 0
G2(0 ) = 0

G3(0 ) = 0 (23)
G~(0) = 0 (24)
G'{(0) = o (2S) K3(0) = 1 (26)

Cr(0)= o Gi(0)-0
G(~'(0) = 0
K0 (0)= 0

G,~(0)= 1 G~(0)= 0
K2(0) = 0

G~'(0)= 0
KI(0) = 0

Now that all initial conditions are spedfied, the eight equations for F 0, F1, F2, F3, GO, G1, G2, and G 3 can be integrated using a Runge-Kutta scheme in one G " + 3 K - S[2G'+ r/G" + g ' G ' - 21-G" (f+ g ) l g ' ( F + G ) ] sweep without any iteration. The terminal values (at = 1) ofF0, F1, F2, F 3 and GO, G1, G 2 and G 3 and = - {go" + 3k0 -S[2g~ + ng 0 + 1 (g0)2 _21g0" (f0 + g0)l} their derivatives can now be used in the boundary condition at 7?= 1, that is, equation (14), to give three (11) simultaneous algebraic equations. The solution gives a, b and c as follows. K'= 0 . (12) a=Aa/A, b=Ab/A, c=Ac/A (27) F (0) = G(0)= F"(0)= G"(0) = 0 (13)
F'(1) = G'(1) = 0, F(1)'+ G(1)= 0 (14)

= - {f;" + k0 - S{2f0 + ~fo + 1(f0)2 -2Lf0(f0 + go)]} (10)

where F n= d (fn), G n= d dr dr (gn), K= d__kk dr (15)

where the determinants are - Fr(I ) F~(1)

V~(1) %(i)
116

I(28)

-[Fo(I) + GO(1)1 F2(1) + G2(1) F3(1) + G3(1 )

Journal of Computational and Applied Mathematics, volume 7, no. 2, 1981.

Fi(1 ab= Gi(1)

-F6(1 ) -G6(1)

F~(1) G~(1)
F3(1)+G3 (1) (29)

tFl(1)+Gl(1)-[F0(1)+G0(1)]

icaUy spaced values of S and fl covering the entire range of practical interest. However, the detailed solutions for fand g have to be excluded due to space limitations but the authors, if requested, would be pleased to furnish them to the readers.

Fi(1 ) F~(1) -F6(1 ) 5. DISCUSSION Ac= Gi(1 ) a~(1) -G(~(1) A continuation method has been developed to solve the nonlinear eigenvalue problem describing the squeezFI(1)+ G1(1) F2(1)+G2(1) -[F0(1)+G0(1)I ing flow of a viscous fluid between elliptic plates. The
attractive features are that the method is conceptually simple and dispenses with the usual iteration. In a Fi(1 F~(1) F~(1) previous numerical study of the problem [3], a new homotopy method had to be used for S < 20, and a A= Gi(1 ) G~(1) G~(1) standard imbedding technique for S > 20. In contrast, F1(1)+Gl(1) F2(1)+G2(1) F3(1)+G3(1) the present scheme worked for all values of S and gave results of the same accuracy in a simpler and more efficient manner. (31) One striking feature of this method is that the method (4) Once a, b, c are known, F, G and K and their derivais insensitive to the first approximation. In fact, the tives can be calculated from equations (16-18). same first approximation is chosen for the solutions (5) We now integrate equation (15) with respect to r for all values of S and ft. To demonstrate, an inspecto give the values at r = Ar as tion of equation (6) shows that the first approximation of the eigenvalue k is chosen to be k 0 = 1 for all (32) fnlr=Ar=fn]r=O+ Fnlr=O.Ar cases. However, the same first approximation proceeded to the final solution automatically in each case. n n The most severe case is the one shown in table 1, where (33) g Ir=ar=g Ir=0+Gntr=0 'At the final solution is k = 173.1767. With iterative methods, one can realize the difficulties in obtaining k i t = a t = k i t = 0 + KIr=0"a (34) the required solution by starting from a first approximation, which is 1/173 o f its final value. (6) To obtain fn, gn and k at r = 2At, the functions It is interesting to observe that for S < 0, which corf, f', f" in equation (10) and the functions g, g" and responds to the case of separating plates, the homotopy g"in equation (11) are replaced by their new values method [3] gave two solutions, one with positive eigenfrom step 5. Steps 3, 4 and 5 are repeated. value and the other with negative eigenvalue. However, (7) The process is continued until r = 1. The functhe perturbation method [3] gave only one solution tional values at r = 1 are the desired solutions of equawith positive eigenvalue. The Present scheme falls in tions (1-4). line with the perturbation solution in the sense that it too produced one solution with positive eigenvalue. In all probability, the negative eigenvalue gives rise to 4. NUMERICAL RESULTS an unstable solution, which is not likely to occur in practice. However, a stability analysis is needed to Numerical results were generated for a range of values resolve the situation completely. of S and ft. The step sizes used in the computation were ~ = 0.002 and Ar = 0.05. A detailed sample solution for S = 25 and fl = 0 is given in table 1. It 6. ACKNOWLEDGEMENT shows the progress of the solution as r changes from 0 to 1. The present value of k = 173.1767 compares The first author is grateful to his wife and children, extremely well with the values of 175.34 and 173.95 who permitted him to spend the entire summer of obtained respectively from imbedding technique and 1980 away from them, to carry out this work. matched asymptotic expansions [3]. Table 2 compares the present values of k with those reported in [3] for specific values offl and S. Note REFERENCES that the present results are quoted only for cases 1. C A M E R O N A. : Principles of lubrication, John Wiley,New where the corresponding data from [3] are available. York, 1966. As can be seen, the agreement is good for all values of S andfl. 2. GRIMMR. J. : "Squeezing flows of Newtonian liquid films, an analysis including fluid inertia", Appl. Sci. Res. 32, 149Since the data presented in [3] are limited, we 166 (1976). present (table 3) additional data for k for systematJournal of Computational and Applied Mathematics, volume 7, no. 2,1981. 117

(30)

3. WANG C. Y. and WATSON L. T. : "Squeezing of a viscous fluid between eMptic plates", Appl. ScL Res. 35,195-207, 1979. 4. NAT. Y. and KURAJIAN G. M. : " A boundary-value solution of the problem of a liquid-sealing circular plate with

f'mite deflections", Industrial Mathematics 29,105-114, 1979. 5. N A T . Y. : Computational methods in engineering boundary value problems, Academic Press, New York, 1979.

T A B L E 1. S a m p l e s o l u t i o n for [3 = 0, S = 2 5 r 0.00 17 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0~6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0 1.0 . f 0.0000 0.3090 0.5878 0.8090 0.9511 1.0000 0.0000 0.3315 0.6422 0.9111 1.1144 1.2054 0.0000 0.3605 0.7081 1.0284 1.2960 1.4318 0.0000 0.3975 0.7891 1.1658 1.5024 1.6869 0.0000 0.4456 0.8910 1.3325 1.7459 1.9848 f" 1.5708 1.4939 1.2708 0.9233 0.4854 0.0000 1.6751 1.6227 1.4666 1.2040 0.7955 0,0000 1.8129 1.7813 1.6838 1.5000 1.1230 0.0000 1.9923 1.9780 1.9309 1.8193 1,4752 0.0000 2.2282 2.2281 2.2235 2.1772 1.8661 0.0000 f" 0.0000 - 0.7625 1.4503 - 1.9962 - 2.3467 - 2.4674 0.0000 0.5224 1.0387 1.6113 2.6225 6.2219 g 0.0000 0.3090 0.5878 0.8090 0.9511 1.0000 0.0000 0.2461 0.46820.6440 0.7563 0.7946 0.0000 0.1766 0.3359 0.4617 0.5413 0.5682 0.0000 0.0980 0.1862 0.2554 0.2988 0.3131 0.0000 0.0054 0.0101 0.0134 0.0150 0.0152 g" 1.5708 1.4939 1.2708 0.9233 0.4854 0.0000 1.2511 1.1900 1.0117 0.7324 0.3816 0.0000 0.8980 0.8539 0.7249 0.5221 0.2690 0.0000 0.4983 0.4734 0.4005 0.2860 0.1447 0.0000 0.0279 0.0258 0.0203 0.0122 0.0040 0.0000 g" 0.0000 0.7625 1.4503 1.9962 2.3467 2.4674 0.0000 0.6070 1.1629 1.6056 1.8669 1.9201 0.0000 0.4380 0.8432 1.1652 1.3355 1.3358 k 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 37.2379 37.2379 37.2379 37.2379 37.2379 37.2379 77.1069 77.1069 77.1069 77.1069 77.1069 77.1069 121.7197 121.7197 121.7197 121.7197 121.7197 121.7197 173.1767 173.1767 173.1767 173.1767 173.1767 173.1767

0.25

0.50

0.0000 - 0.3174 - 0.6707 - 1.2364 - 2~816 -10.2820 0.0000 - 0.1448 - 0.3466 - 0.8758 - 3.1303 -14.7493 0.0000 - 0.0024 - 0.0667 - 0.5334 - 3.3793 -19.8179

0.75

0.0000 -0.2474 - 0.4767 - 0.6564 - 0.7351 - 0.7031 0.0000 0.0197 0.0351 0.0436 0.0347 .02022

1.00

J o u r n a l o f C o m p u t a t i o n a l a n d A p p l i e d M a t h e m a t i c s , v o l u m e 7, n o . 2, 1 9 8 1 .

118

TABLE 2. Comparison of values of k. PM = present method, HM = homotopy method [3], RP = regular perturbation [3] IM = imbedding method [3], MAE = matched asymptotic expansions [3] S= - 0 . 5 PM 0 0.2 0.25 0.50 0.70 1.00 HM RP PM S= 0 HM RP PM S= 1.0 HM RP PM S= 25.0 IM MAE

1.8099 1.7865 1.7982 6.0063 1.7683 1.7527 1.8330 4.8050 1.6311 1.6248 1.6662 4.0104 1.5048 1.5018 1.5163 3.5386 1.3036 1.3023 1.3045 3.0078

6.0000 6.0000 13~8984 13.9443 13.8785 173.1767 175.34 173.95 4.8000 4.0000 3.5294 3.0000 4.8000 10.3809 10.3699 10.2472 4.0000 8.4530 8.4336 8.3791 3.5294 3.0000 6.2778 6.2603 6.2465

99.8439 99.979 98.401 87.1966 87.261 85.850 74.0414 73.865 72.655

TABLE 3. Additional data for k. 0 -0.5 0 0.5 1 5 10 15 25 1.8099 6.0063 10.0227 13.8 984 42.6763 76.4796 109.3530 173.1767 0.2 1.7683 5.0052 8.0075 10.9030 32.5195 57.9693 82.7145 131.1877 0.4 1.6857 4.2902 6.7409 9.1165 26.9326 47.9310 68.3460 108.3300 0.6 1.5701 3.7539 5.8510 7.8 904 23.2204 41.2975 58.8715 93.2887 0.8 1.4374 3.3368 5.1839 6.9829 20.5186 36.4829 52.0027 82.3960 1.0 1.3036 3.0078 4.6618 6.2778 18.4401 32.7854 46.7310 74.0414

Journal of Computational and Applied Mathematics, volume 7, no. 2, 1981.

119