Numerical Heat Transfer, Part B, 38:133^156 , 2000 Copyright # 2000 Taylor & Francis 1040-7790 /00 $12.00 + .00 AN EFFIC IENT M ETH OD OF COM PUTING EIGENVALUES IN HEAT COND UC TION A. Haji-Sheikh Department of Mechanical and Aerospace Engineering, The University of Texas at Arlington, Arlington, Texas 76019-0023 , USA J. V. Beck Department of Mechanical Engineering, Michigan State University, East Lansing, Michigan 48824-1226 , USA Ef cient algorithms for computing eigenvalues for heat conduction problems in Cartesian and spherical coordinates are given. Explicit approximat e relations are presented that generally provide accurate results. When these approximate relations are followed by a high-order Newton root- nding iteration, a high degree of accuracy can be realized. It is demonstrated that in Cartesian coordinates, eigenvalues with excellent accuracy are obtained over the entire range of parameters. INTROD UCTION In this era of powerful computers and well-developed numerical methods, a need still exists for exact solutions in heat conduction. These solutions can be used to aid in the veri¢cation of complex numerical programs; to provide insight into various heat conduction problems; and to aid students’ understanding of some steady-state and transient heat transfer phenomena. In the veri¢cation of programs, particularly important are extremely accurate solutions for two- and threedimensional transient problems. These accurate solutions can, in turn, require many terms in the double or triple series and also accurate eigenvalues. However, obtaining extremely accurate eigenvalues can be awkward or inef¢cient for the convective boundary condition. The objective of this article is to provide means of obtaining these eigenvalues in a manner which itself has been veri¢ed for a wide range of Biot numbers, a great many eigenvalues, and for several combinations of convective conditions. The eigenvalues for the convective boundary conditions are intended for one-, two-, or three-dimensional Cartesian coordinates in transient cases. For steady-state heat conduction, two- and three-dimensional cases are important. Even though multiple directions are in mind, only one direction need be considered here for the eigenvalues, because the solutions are frequently obtained using products of one-dimensional components or Green’s functionsösee Beck et al. [1]. Only homoReceived 22 December 1999; accepted 4 February 2000. Address correspondence to Dr. A. Haji-Sheikh, Department of Mechanical and Aerospace Engineering, P.O. Box 19023 UTA Station, Arlington, TX 76019-0023 , USA. E-mail: haji@mae.uta.edu 133 134 A. HAJI-SHEIKH AND J. V. BECK NOM ENCLATUR E Am Bi cn dn Eij h H k L n p S Xij constants Biot number ( ˆ hL/k) (n ¡ 34)p (n ¡ 12)p function of Bi 1 and Bi 2 heat transfer coef¢cient, W/m2 K function of n thermal conductivity, W/m K thickness, m eigenvalue index 1, 2, etc. constant constant Cartesian x-coordinate with boundary conditions of the ith and jth kinds zn bn gn;ij En zn xij nth eigenvalue approximation exact nth eigenvalue coef¢cient deviation zn ¡ zn nth eigenvalue, estimated function Subscripts 1 boundary conditions of the ¢rst kind 2 boundary conditions of the second kind 3 boundary conditions of the third kind geneous bodies with temperature-independent properties are usually implied when these eigenvalues are used. Three sets of eigenconditions for Cartesian coordinates are considered, each of which has a different set of eigenvalues, although the third condition to be considered can include the ¢rst two. Incidentally, one of the eigenconditions also applies for a solid sphere with a convective boundary condition at its surface. To describe the problems, it is convenient to use some of the notation in [1]. This notation builds on the common descriptions of the boundary conditions as being of the ¢rst kind (that is, prescribed temperature), second kind (prescribed heat £ux), and third kind (convective condition with a prescribed ambient temperature). A plate which has a prescribed temperature history at x ˆ 0 and a convective boundary condition at x ˆ L is denoted by X13. A plate which has a prescribed heat £ux at x ˆ 0 and a convective boundary condition at x ˆ L is denoted by X23. Each has a particular eigencondition in terms of a transcendental equation, which is the same for X13 and X31 and also is the same for X23 and X32. The X13 eigencondition and eigenvalues are nearly the same as for a solid sphere with a convective boundary condition (which is denoted by RS03 in [1]). Actually, several other spherical cases use the same eigencondition, namely, RS12, RS13, RS21, and RS31. The X13 eigencondition is also the same as for the X42 (or X24) case, which has boundary conditions of the second and fourth kinds ( a high-conductivity ¢lm at the surface); see Beck et al. [1]. The third kind of boundary condition is denoted by X33, which has a convective boundary condition on both surfaces. The heat transfer coef¢cients may or may not be the same on both sides. Actually, this case includes the other two because the heat transfer coef¢cient going to zero gives the boundary condition of the second kind and the heat transfer coef¢cient going to in¢nity gives a boundary condition of the ¢rst kind. This X33 case is the most dif¢cult; eigenvalues have been tabulated and simpli¢ed equations have been given in the literature for the X13 and X23 cases, but the X33 case has rarely been treated. This set of eigenvalues can also be used for the RS22, RS23, RS32, and RS33 cases. A brief literature review is now given. A number of ways have been proposed to obtain numerical values for the eigenvalues for the X13, X23, and X33 cases in heat EIGENVALUES IN HEAT CONDUCTION 135 condition. Values for the X13 and X23 cases are tabulated in many books, such as Zucker [2]. Yovanovich [3] has proposed simple algebraic equations for the ¢rst eigenvalues for the cases of X13, X23, and R03 (solid cylinder with a convective boundary condition). The maximum errors are about 0.002, which is quite good but not good enough for extremely accurate computations. No accurate equations are given for the higher eigenvalues. Beck et al. [1] has proposed some algebraic equations for the X13, X23, and X33 cases. These are based on series expansions. All the eigenvalues are considered, but sometimes multiple equations are necessary for a given case and the entire range of the Biot numbers is not covered for the X33 case. However, a method of obtaining extremely accurate values is given for the ranges where the approximations are valid. Another procedure for the explicit calculation of the eigenvalues for the X23 and X13 cases is given by Leathers and McCormick [4]. However, the computation involves numerical integration, possibly needing on the order of 105 points. Numerous techniques for computing eigenvalues are described in [5^7]. Aviles-Ramos et al. [8] introduce a hybrid search algorithm based on central differencing that can be widely used without explicit differentiation of the transcendental equations. Recently, Stevens and Luck [9] have proposed attractive sets of equations. These equations are attractive because they are ef¢cient and can be made as accurate as desired, consistent with the number of signi¢cant ¢gures used in the computation. (These sets of equations can be put in the form of algorithms for ef¢cient programming.) The method of Stevens and Luck [9] for the X23 case, that is, one convective boundary and one insulated surface, is modi¢ed herein to achieve even faster convergence. For a Dirichlet boundary condition at one surface and a convective boundary condition at the other surface, the X13 case, two algorithms are presented in [9], one for the ¢rst eigenvalue and another for all the others. Herein a single accurate algorithm is given covering all the eigenvalues. These two algorithms (for the X13 and X23 cases) are properly combined for application to problems when both surfaces are subject to convective boundary conditions, the X33 case. Stevens and Luck do not consider the dif¢cult X33 case. The main objective of this study is to investigate a robust method of ¢nding extremely accurate eigenvalues for one-dimensional transient problems with convective boundary conditions. (As pointed out above, these same eigenvalues are used in two- and three-dimensional problems in Cartesian coordinates and some problems in spherical coordinates.) This objective leads to modi¢cation and extension of Stevens and Luck [9] to meet the following self-imposed goals: 1. 2. 3. 4. The methods should provide an explicit equation with, at least, six-decimal place accuracy, exceeding typical accuracy of tabulated information. The method must be suf¢ciently robust that there is no numerical instability within the computer word limitation. The basic equations must have suf¢ciently simple forms. The methods must be capable of being expressed in terms of algorithms that can be quickly programmed using a variety of computer languages. To meet these goals, a preliminary function that approximates the eigenvalues is obtained based solely on the asymptotic behavior of eigenfunctions for the X13, X23, and X33 cases. An accurate estimate is an important initial step for a quick 136 A. HAJI-SHEIKH AND J. V. BECK evaluation of eigenvalues. Later, the results of this study are modi¢ed to accommodate one-dimensional radial conduction in spherical bodies. An outline of the remainder of this article is now given. The algorithm for the X23 case is ¢rst developed. One of the most important parts of this research is to demonstrate a methodology for ¢nding the algorithms for the eigenvalues; it is illustrated in this section and in the next one, which addresses the X13 case. In the following section, the results of these two sections are used to develop an algorithm for the X33 case. Finally, the application to the RS03 (and other cases) is discussed by introducing some minor modi¢cations. ANALYSIS X23 or X32 Eigenvalues The transcendental equation for the X23 and X32 cases is bn tan bn ¡ Bi ˆ 0 …1† where bn is the nth eigenvalue. The eigenvalues for this case have the following limiting conditions: As Bi ! 0 As Bi ! 1 bn ! …n ¡ 1†p bn ! n ¡ 12 p which should be strictly adhered to. An algorithm based on the article of Stevens and Luck is brie£y described here. A function that provides approximate eigenvalues, as given in [9], is p zn ˆ cn ‡ x23 …2a† 4 where and cn ˆ …n ¡ 34†p …2b† Bi ¡ cn Bi ‡ cn …2c† x23 ˆ satis¢es the limiting conditions as described above. Equation (2a) has another remarkable feature: it satis¢es the condition suggested by Eq. (1), bn ˆ Bi, when tan bn ˆ 1. Also, according to Eqs. (2a)^2(c), when Bi ˆ cn , then zn ˆ cn ˆ Bi, and using zn instead of bn , one can satisfy Eq. (1). However, the function zn does not satisfy the asymptotic behavior of bn as Bi!0, described in [9]. For a set of computed eigenvalues to be complete, it is essential to have highly accurate zn as it approaches (n¡1)p and (n¡12)p. Equation (2a) should be modi¢ed so that its slope of zn becomes the same as dbn =d Bi as Bi! 0. The value of dbn =dBi is computed by differentiating Bi in the eigenfunction relation, Eq. (1), with respect to EIGENVALUES IN HEAT CONDUCTION 137 bn ; that is, dbn 1 ˆ dBi bn ‡ …1 ‡ Bi† tan bn …3a† Since bn ! …n ¡ 1†p as Bi! 0, then tan bn ˆ 0 and dbn 1 ˆ dBi …n ¡ 1†p When n > 1 …3b† Moreover, when n ˆ 1, Eq. (1) suggests tan bn ! bn as Bi!0, and that produces p bn ˆ Bi. The plan is to include this asymptotic behavior of bn when Bi approaches zero for n ˆ 1 and the limiting slope given by Eq. (3b) in the formulation of zn . In order to maintain the form of the function x23 , as suggested in [9], and incorporate this asymptotic behavior of bn , Eq. (2a) is modi¢ed to take the following form: p zn ˆ cn ‡ x23 gn;23 4 …4† The parameter gn;23 in Eq. (4) must satisfy the following conditions: (a) Have unit values both as Bi!0 and Bi! 1 (b) Maintain zero slope for p zn as Bi!1 (c) Force zn to approach * Bi, when n ˆ 1, as Bi!0 (d) Provide dzn =dBi ^1/[p(n¡1)], when n > 1, as Bi!0 The factor gn;23 is selected to perform these tasks, without altering the characteristics of Eq. (2a), as gn;23 ˆ 1 ¡ 1:04 Bi ‡ cn ¡ p=4 Bi ‡ H…cn ¡ p=4† ¡ Bi ‡ H Bi ‡ H …5† Equation (4) with gn;23 from Eq. (5) satis¢es conditions (a) and (b) unconditionally, and it can produce in¢nite slope only for n ˆ 1. When H ˆ …p=4†2 , condition (c) is satis¢ed; however, Eq. (5), in its present functional form, does not satisfy condition (d) for n > 1 explicitly. Using the H value for n ˆ 1 resulted in signi¢cant improvement of the maximum deviation, bn ¡ zn , to 0.0049 in comparison with 0.0531 when gn;23 ˆ 1. This motivated a search for an alternative functional form to be used instead of Eq. (4) while maintaining the de¢nition of z23 , Eq. (2c). An alternative functional form of zn is sought that would permit satisfying condition (d). Among the possible candidates, a linear combination of two functions, z23 and tanh(z23 )/tanh(1), is used as summarized in the equation zn ˆ cn ‡ pgn;23 ¡1 tanh…x23 † n x23 ‡ …1 ¡ n¡1 † 4 tanh…1† …6† To satisfy condition (d) using Eq. (6), the function describing the limit of dz n =dBi, as Bi! 0, is set equal to 1/[p(n¡1)]; that is, replacing dbn =dBi in Eq. (3b) by dz n =dBi. Then H, the root of the resulting transcendental equation, is numerically computed for a selected number of n values. The solution is sensitive to the value of S ˆ 1:04 and the best results are obtained when S has near-unit values. The numerical computation of H begins using a constant S ˆ 1 instead of 1.04 in Eq. (5). 138 A. HAJI-SHEIKH AND J. V. BECK The discrete data in Figure 1 are the computed values of H. A linear function, H ^ 0:75 ‡ 1:43…n ¡ 1† …7a† In Figure 1 is the least-squares approximation of the discrete data. This signi¢cantly improves the accuracy of zn , especially at small Bi values; however, it is not the optimum value of H when n is small, e.g., n ˆ 1. Slight adjustments are made to improve the accuracy of zn for small n values, and the equation H ^ 0:76 ‡ 1:22…n ¡ 1† …7b† with S ˆ 1.04 performed well. These latter values of S and H are used for the recommended algorithm for the X23 case. Figure 2 demonstrates the suitability of zn over a broad range of Biot numbers. The values of bn ¡ zn in Figure 2 describe the relative error of zn . Figure 2a is for gn;23 ˆ 1 in Eq. (4), which corresponds to zn in Eq. (2a), the Stevens and Luck [9] equation. Next, the function zn in Eq. (6) serves as an initial estimate of the root of Eq. (1). Figure 2b is prepared using Eq. (6) with Eq. (5), for gn;23 , H from Eq. (7b), and S ˆ 1.04. A modi¢ed second-order Newton method, described in the Appendix, provides the value of zn , the ¢rst approximation for bn . The estimated difference between bn Figure 1. Estimated values of H in the de¢nition of gn;23 , Eq. (5). EIGENVALUES IN HEAT CONDUCTION 139 Figure 2. The bn ¡ zn deviation as a function of Bi for X23 or X32 case when n ˆ 1; 2; 5; 10; 100: (a) Eq. (2a); (b) Eq. (6). and zn is En , obtained from the relations A0 ˆ Bi cos zn ‡ zn sin zn …8a† A2 ˆ cos zn …8c† A1 ˆ …1 ‡ Bi† sin zn ‡ zn cos zn en ^ ¡ A0 A1 1‡ …A2 =A1 †…A0 =A1 † 1 ¡ 2…A2 =A1 †…A0 =A1 † …8b† …8d† Equations (8a)^(8c) are the same as those in [9]. A strict second-order Newton method formulation contains an additional ¡A0 /2 term on the right-hand side of Eq. (8c). However, dropping ¡A0 /2 improves the results, assuming A0 ˆ 0 if zn ˆ bn in the de¢nition of eigencondition, Eq. (1). The deviation given by Eq. (8d) yields the approximate root of a quadratic equation, A0 ‡ A1 E ‡ A2 E2 ˆ 0. Then, 140 A. HAJI-SHEIKH AND J. V. BECK the eigenvalue zn for the X23 case is zn ˆ zn ‡ e n …9† An algorithm describing calculation of eigenvalues for X23 is given in Table 1. A computer program can be written using these equations in the order given. To compute zn , one calculates zn for inclusion in Eqs. (8a)^(8c) and Eq. (8d) provides En . Figure 3 shows the accuracy of zn after one iteration. The deviations of zn from the true eigenvalues, when using Eq. (2a), for different n and Bi values are in Figure 3a. This is similar to the results reported in [9]. Figure 3b is for when zn is taken from Eq. (6). Figure 3b shows the computed deviation of zn from the exact eigenvalue bn , that is, the expected error bn ¡ zn , over a broader range of Biot numbers. Notice that the data in Figure 3b have much lower errors than the data in Figure 3a, by a factor of nearly 104 . This is due to a better than one-decimal-place improvement of bn ¡ zn in Figure 2b over Figure 2a. The largest value of n appearing in Figures 2 and 3 is 100; however, the numerical calculations were successfully tested for n over 105 and for all Biot numbers from above 1010 to below 10¡10 . When n > 100, the value of the maximum error stabilizes at *6£10¡8 and the peaks in Figures 3a^3b gradually shift toward higher Bi values. The peaks are located near the crossing points at bn ˆ zn ˆ cn . To demonstrate the accuracy of the eigenvalues numerically, zn is computed over a small range of Biot numbers and for n ˆ 1 through 5; the results are in Table 2. The eigenvalues computed using one and two iterations are computed with data from [2]. For one iteration, the errors in zn data agree with deviations shown in Figure 3b. X13 or X31 Eigenvalues Only a minor modi¢cation of Eqs. (2a)^(2c), that is, changing 34 to 14, can provide a zn relation for the X13 case. This provides acceptable results when n ˆ 1. Table 1. Algorithm for computing the eigenvalues Quantities Select n and the Biot numbers(s) and use equations given below for X13 X23 X33 cn dn x13 x23 H gn;13 gn;23 zn A0 A1 A2 A3 En xn ö (11b) (13a) ö ö (13c) ö (13d) (14a) (14b) (14c) ö (8d) (15) (2b) ö ö (2c) (7b) ö (5) (6) (8a) (8b) (8c) ö (8d) (9) (2b) (11b) (17b) (18b) (7b) (13c) (18c) (20) (21a) (21b) (21c) (21d) (21e) (22) EIGENVALUES IN HEAT CONDUCTION 141 Figure 3. The computed deviation, bn ¡ zn ; as a function of Bi for X23 when n ˆ 1; 2; 5; 10; 100: (a) using zn from Eq. (2a); (b) using zn from Eq. (6). However, without a correct value of dbn =dBi as Bi!0, when n is large, a relatively large error shifts toward Bi ˆ 0. Since the value of dbn =dBi, as Bi! 0, cannot be conveniently incorporated in the de¢nition of zn , an alternative scheme is used. For this or the X31 case, the eigenvalues are the roots of the equation bn cot bn ‡ Bi ˆ 0 This equation has the following limiting values: As Bi ! 0 As Bi ! 1 bn ! …n ¡ 12†p bn ! np …10† 142 A. HAJI-SHEIKH AND J. V. BECK Table 2. Calculated eigenvalue, zn , for X23 using explicit approximate formulation and comparison with results after two and three iterations n 1 2 3 4 5 Bi 0.1 1.0 10 100 0.1 1.0 10 100 0.1 1.0 10 100 0.1 1.0 10 100 0.1 1.0 10 100 zn , Eq. (4) zn Zucker [2] After two iterations 0.31832 0.86127 1.43135 1.55541 3.17316 3.42210 4.30817 4.66492 6.30003 6.44124 7.22828 7.77558 9.43622 9.53408 10.20029 10.88711 12.57503 12.64980 13.21500 13.99889 0.311052791 0.860333589 1.428870010 1.555245130 3.173097177 3.425618473 4.305801409 4.665765142 6.299059360 6.437298159 7.228109772 7.776374078 9.435375976 9.529334370 10.200262588 10.887130102 12.574323161 12.645287193 13.214185684 13.998089735 0.31105 0.86033 ö ö 3.17310 3.42562 ö ö 6.29906 6.43730 ö ö 9.43538 9.52933 ö ö 12.57432 12.64529 ö ö 0.311052848200298 0.860333589019380 1.428870011214077 1.555245129256167 3.173097176692870 3.425618459481728 4.305801413119223 4.665765141727248 6.299059359895646 6.437298179171947 7.228109771627249 7.776374077846953 9.435375975760847 9.529334405361963 10.20026258829591 10.88713010214771 12.57432316103787 12.64528722385664 13.21418568384292 13.99808973515508 The function p zn ˆ dn ‡ x13 2 …11a† 1 p 2 …11b† Bi Bi ‡ o …11c† where dn ˆ n ¡ and x13 ˆ satis¢es the aforementioned limiting conditions. The variable o is a constant that depends on n, and it is selected so that the slope dz n =dBi is the same as the slope of the exact eigenvalue dbn =dBi as Bi!0. Accordingly, the slope dbn =dBi is computed by differentiating the eigenfunction relation, Eq. (10), with respect to bn to obtain dbn 1 ˆ dBi bn ¡ …1 ‡ Bi† cot bn …12a† As Bi!1, bn ! np, then dbn =dBi!0 and as Bi! 0, one can set bn ˆ …n ¡ 12†p, then EIGENVALUES IN HEAT CONDUCTION 143 cot bn ! 0, and Eq. (12a) reduces to dbn 1 1 ˆ ˆ dBi bn …n ¡ 12†p …12b† Similarly, one can differentiate zn in Eq. (11a) with respect to Bi to obtain dbn p o p ˆ ˆ ˆ 2 dBi 2 2o …Bi ‡ o† Biˆ0 Biˆ0 …12c† and then set p/(2o) ˆ 1/[n¡12]p or o ˆ p2 …n ¡ 12†=2 ˆ pdn /2; this reduces Eq. (11c) to x‰13 ˆ Bi Bi ‡ pdn =2 …13a† The above formulation for zn is based solely on the matching of the asymptotic behavior of zn at small and large Biot numbers. Accordingly, to enhance the accuracy of zn in Eq. (11a), it is reasonable to include some mid-range corrections. To reduce the deviation bn ¡ zn within 0 µ z13 µ 1 without altering the asymptotic behavior of zn , one can set p bn ¡ zn ˆ p1 x213 …1 ¡ x13 † 2 where p1 is a constant for any n. The parameter p1 was computed for a selected number of n values so that …bn ¡ zn †2 is minimum within 0 µ x13 µ 1 and its approximate value is p1 ^ 1 ¡ 0:85=n. The modi¢ed Eq. (11a) is p zn ˆ dn ‡ x13 ‰1 ‡ p1 x13 …1 ¡ x13 †Š 2 …13b† According to Eq. (13b), there will be one additional point at x13 ˆ 0:6 ‡ 0:245=n, between x13 ˆ 0 and 1, where zn % bn . The accuracy of zn in Eq. (13a) is further enhanced by repeating the aforementioned process while using the relation p bn ¡ zn ˆ p2 x213 …1 ¡ x13 †…1 ‡ x13 †…p3 ¡ x13 † 2 where p2 is a constant to be determined by p3 ^ 0:6 ‡ 0:245=n approximates the locations where zn ˆ bn . The method of calculating p2 ^ 0:6 ¡ 0:71=n is the same as that described for p1 . In summary, one can de¢ne a new parameter gn;13 as gn;13 ˆ 1 ‡ x13 …1 ¡ x13 † 1 ¡ 0:85 0:71 0:245 ‡ …0:6 ¡ …1 ‡ x13 † 0:6 ‡ ¡ x13 n n n …13c† and compute zn from the equation p zn ˆ dn ‡ gn;13 x13 2 …13d† The variable zn in Eq. (13a) serves as an initial estimate of the root of Eq. (10). The modi¢ed higher-order Newton method (see Appendix) can provide accurate results if zn is a suitable estimate of the eigenvalues. Accordingly, one can de¢ne En as the difference between bn and zn and estimate En from Eq. (8d) using the following 144 A. HAJI-SHEIKH AND J. V. BECK parameters: A0 ˆ Bi sin zn ‡ zn cos zn …14a† A2 ˆ sin zn ¡ 0:5A0 …14c† A1 ˆ …1 ‡ Bi† cos zn ¡ zn sin zn …14b† According to Eq. (10), one can delete the ¡0.5A0 term in Eq. (14c). Unlike the X23 case, the inclusion of ¡0.5A0 in Eq. (14c) improves the accuracy of the results. Indeed, an intermediate value such as ¡0.3A0 instead of ¡0.5A0 in Eq. (14c) further improves the accuracy. Then, following computation of En using Eq. (8d), the relation zn ˆ zn ‡ e n …15† describes an approximate value of the eigenvalues bn . The computation procedure is identical to that described earlier for the X23 case. A step-by-step algorithm suitable for computer programming is in Table 1. A sample of computed zn , using Eq. (15), is in Table 3. Comparing the last two columns, the computed zn exhibits remarkable accuracy. Therefore, it is appropriate to study the accuracy of the computed eigenvalues bn approximated by zn , over a large range of Biot numbers. To demonstrate the suitability of zn , the value of bn ¡ zn is plotted in Figure 4 over a large range of Biot numbers, Bi. Without using any Newton iteration, Figure 4 yields good accuracy throughout the domain. The computed values of bn ¡ zn are plotted in Figure 5. Figure 5 contains the deviation from the exact value for n ˆ 1, 2, 5, 10, and 100. The solution was stable for Biot numbers between 10¡10 Table 3. Calculated eigenvalue, zn , for X13 using explicit approximate formulation and comparison with results after two and three iterations n Bi zn , Eq. (13a) zn Zucker [2] After two iterations 1 0.1 1.0 10 100 0.1 1.0 10 100 0.1 1.0 10 100 0.1 1.0 10 100 0.1 1.0 10 100 1.63211 2.03036 2.85882 3.11016 4.73353 4.91387 5.75469 6.21974 7.86671 7.97955 8.70338 9.33082 11.00467 11.08625 11.69973 12.44245 14.14424 14.20798 14.73283 15.55438 1.6319945272 2.0287578382 2.8627725878 3.1104977023 4.7335118024 4.9131804394 5.7605579348 6.2210548279 7.8666927716 7.9786657124 8.7083138327 9.3317301257 11.004661096 0 11.085538406 5 11.702678081 2 12.442581015 9 14.144236840 7 14.207436725 2 14.733472342 3 15.553662970 8 1.63199 2.02876 ö ö 4.73351 4.91318 ö ö 7.86669 7.97867 ö ö 11.00466 11.08554 ö ö 14.14424 14.20744 ö ö 1.631994527214800 2.028757838110434 2.862772587515207 3.110497702305585 4.733511802356786 4.913180439434884 5.760557932709097 6.221054827821945 7.866692771561574 7.978665712413241 8.708313830875857 9.331730125693797 11.00466109603352 11.08553840649702 11.70267808065236 12.44258101585997 14.14423684071843 14.20743672519119 14.73347234227069 15.55366297078455 2 3 4 5 EIGENVALUES IN HEAT CONDUCTION Figure 4. The bn ¡ zn deviation as a function of Bi for X13 when n ˆ 1; 2; 5; 10; 100: Figure 5. The bn ¡ zn deviation as a function of Bi for X13 when n ˆ 1; 2; 5; 10; 100: 145 146 A. HAJI-SHEIKH AND J. V. BECK and 1010 , well below and above those plotted in the ¢gures. Based on data in Table 3 and data in Figure 5, Eq. (15) can provide eigenvalues with eight accurate decimal places and the accuracy in the mid-range, where the error is the largest, satis¢es the need of critical computations. X33 Eigenvalues It is remarkable that the estimated form of eigenfunctions for the X13 and X23 cases may be combined in a simple manner to construct a working relation for the X33 case. The eigenvalues are the roots of the transcendental equation …b2n ¡ Bi1 Bi2 † tan bn ¡ …Bi1 ‡ Bi2 †bn ˆ 0 The contribution of the X13 case is modi¢ed and will be designated as p E13 ˆ x13 gn;13 2 …16† …17a† where Bi1 Bi2 Bi1 Bi2 ‡ 0:2 ‡ …Bi1 ‡ Bi2 †pdn =2 …17b† pg n;23 ¡1 tanh…x23 † n x23 ‡ …1 ¡ n¡1 † 4 tanh…1† …18a† Bi1 ‡ Bi2 ¡ cn Bi1 ‡ Bi2 ‡ cn …18b† x13 ˆ and gn;13 is given by Eq. (13c). The parameter x13 is de¢ned so that E13 , Eq. (17a), reduced to px13 /2 in Eq. (13d) as Bi1 or Bi2 becomes in¢nite. Similarly, the modi¢ed contribution of the X23 case is E23 ˆ where x23 ˆ and gn;23 ˆ 1 ¡ 1:04 Bi1 ‡ Bi2 ‡ Bi1 Bi2 ‡ cn ¡ p=4 Bi1 ‡ Bi2 ‡ Bi1 Bi2 ‡ H Bi1 ‡ Bi2 ‡ Bi1 Bi2 ‡ H…cn ¡ p=4† ¡ Bi1 ‡ Bi2 ‡ Bi1 Bi2 ‡ H …18c† satis¢es the asymptotic behaviors of zn as Bi1 and/or Bi2 approaches zero; and the condition bn ^(Bi 1 ‡ Bi2 ‡ Bi1 Bi2 ) 1=2 when bn ! 0 is also satis¢ed. One simple method of ¢nding an initial estimate of the eigenvalues is to set a Taylor series expansion of zn …x13 , x23 ), about (x13 , x23 ) equal to (0, 0), to get zn ‡ cn ‡ E13 ‡ E23 …19† The substitution of E13 from Eq. (17) and E23 from Eq. (18a) in Eq. (19) ensures that, as Bi1 or Bi2 approaches in¢nity, Eq. (19) reduces to Eq. (13d). Also, as Bi1 or Bi2 approaches zero, Eq. (19) reduces to Eq. (6). Although Eq. (19) can provide accurate EIGENVALUES IN HEAT CONDUCTION 147 results after a few iterations, the accuracy of zn over a broad range of Bi1 , Bi2 , and n values was below the targeted accuracy of 10¡6 . There are other methods of combining E13 and E23 to construct a zn function. The approximate values of zn for the X13 and X23 cases, despite their high degree of accuracy, only affect the limiting values of zn for X33. Accordingly, a simple method is sought that provides an alternative formulation of zn for the X33 case. This new empirical relation is Bi1 Bi2 Bi1 Bi2 ‡ …cn ‡ E23 † 1 ¡ 2 Bi1 Bi2 ‡ dn Bi1 Bi2 ‡ dn2 Bi1 Bi2 Bi1 Bi2 p p ˆ dn ‡ x13 ; gn;13 ‡ cn ‡ x23 gn;23 1 ¡ 2 2 Bi1 Bi2 ‡ dn 4 Bi1 Bi2 ‡ dn2 zn ˆ …dn ‡ E13 † …20† The coef¢cient dn2 in the denominator is a free numerical constant. It is evaluated by linear regression mainly to reduce the maximum errors for different n values. It is remarkable that the computed constant for different n values closely approximated dn2 . Equations (19) and (20) are presented to show the method combining E13 and E23 is not unique. It is possible to construct other relations; however, the simplicity of the relation and relative accuracy of zn are the primary reasons for selecting Eq. (20) for this numerical presentation. Equation (20) provides an initial estimate of zn , and Figure 6 shows a three-dimensional plot for b1 ¡ z1 . Accordingly, the reasonably good accuracy of data in Figure 6 and a high-order Newton yield eigenvalues with reasonably good Figure 6. The bn ¡ zn deviation as a function of Bi1 and Bi 2 when n ˆ 1 for the X33 case. 148 A. HAJI-SHEIKH AND J. V. BECK accuracy for all combinations of Bi1 , Bi2 , and n. The procedure described for the X13 and X23 cases will be repeated for this X33 study. The second-order Newton’s method was extended to become a third-order modi¢ed Newton method and was applied to this case. The working relations are A0 ˆ …z2n ¡ Bi1 Bi2 † sin zn ¡ …Bi1 ‡ Bi2 †zn cos zn A1 ˆ …2 ‡ Bi1 ‡ Bi2 †zn sin zn ‡ ‰z2n ¡ …Bi1 ‡ Bi2 † ¡ Bi1 Bi1 Š cos zn A0 A2 ˆ …1 ‡ Bi1 ‡ Bi2 † sin zn ‡ 2zn cos zn ¡ 2 2 A3 ˆ f‰6 ‡ 3…Bi1 ‡ 2† ‡ Bi1 Bi2 ¡ zn Š cos zn ¡ …6 ‡ Bi1 ‡ Bi2 †zn sin zn g=6 en ˆ A0 A1 1‡ 2 …A0 =A1 †…A2 =A1 † ¡ …A0 =A1 † …A3 =A1 † 1 ¡ 2…A0 =A1 †…A2 =A1 † ‡ 3…A0 =A1 †2 …A3 =A1 † …21a† …21b† …21c† …21d† …21e† where En approximates the proper root of equation A0 ‡ A1 En ‡ A2 E2n ‡ A3 E3n ˆ 0. As for the X13 and X23 cases, the relation zn ˆ zn ‡ e n …22† provides zn , the approximate eigenvalue. Equation (22) performed well for all n values. Figure 6 is a three-dimensional plot describing the variation of bn ¡ zn for n ˆ 1 as a function of Bi1 and Bi2 . Figure 7 is a similar three-dimensional plot for n ˆ 1, and it demonstrates the value of bn ¡ zn that represents the error in z1 over a broad range of Bi1 and Bi2 following Figure 7. The bn ¡ zn deviation as a function of Bi1 and Bi2 when n ˆ 1. EIGENVALUES IN HEAT CONDUCTION Figure 8. The b2 ¡ z2 deviation as a function of Bi1 and Bi2 for the X33 case. Figure 9. The b2 ¡ z2 deviation as a function of Bi1 and Bi2 for the X33 case. 149 150 A. HAJI-SHEIKH AND J. V. BECK Table 4. Calculated eigenvalue, z1 , for X33 using explicit approximate formulation and comparison with results after two and three iterations Bi1 Bi2 zn , Eq. (21) z1 After two iterations After three iterations 0.1 0.1 1.0 10 100 1.0 10 100 10 100 100 0.44420 0.92168 1.49185 1.61693 1.30592 1.88961 2.01550 2.62089 2.83147 3.07887 0.443520787 9 0.929253116 8 1.489910836 6 1.616419903 8 1.306542374 2 1.875307809 5 2.011948778 4 2.627675433 0 2.836813028 3 3.080011883 8 0.44352078788188 8 0.92925310992520 6 1.48991083663246 6 1.61641990379493 8 1.30654237418880 6 1.87530781059643 6 2.01194877843164 1 2.62767543298579 7 2.83681302827527 7 3.08001188380088 4 0.44352078788188 9 0.92925310992520 6 1.48991083663246 6 1.61641990379493 8 1.30654237418880 6 1.87530781059643 6 2.01194877843164 1 2.62767543298579 7 2.83681302827527 7 3.08001188380088 4 1.0 10 100 one iteration. It is observed from the ¢gures that the relations presented above satisfy the asymptotic behavior of bn . Based on data plotted in Figure 7, there are better than six accurate decimal places over a broad range of the Biot numbers when n ˆ 1. Figure 8 demonstrates the value of bn ¡ zn as a function of Bi1 and Bi2 when n ˆ 2. Figure 9 is a similar three-dimensional plot describing the variation of bn ¡ zn when n ˆ 2. Figure 9 shows relatively smaller errors than those in Figure 7. Indeed, when n > 1, Eqs. (20)^(22) yield eigenvalues with 7 or more accurate decimal places. The error in zn is largest when n ˆ 1, and Table 4 demonstrates the accuracy of z1 over a range of Bi1 and Bi2 . The data show excellent accuracy, especially when the Biot number is large. Examining the three-dimensional graphs for higher n values, up to n ˆ 104 , shows that the maximum error remains less than 10¡7 . Moreover, it is possible to increase the accuracy of computed bn using a second iteration. This is demonstrated by plotting the errors in computed bn for n ˆ 1 and 2, and the results are in Figures 10 and 11, respectively. This indicates that the accuracy of data in Figures 7 and 9 can be greatly enhanced by an additional iteration. C ONC LUSION AND REM ARK S Extremely accurate forward equations are given for the basic eigenconditions in transient heat conduction for convective boundary conditionsönamely, for the boundary conditions denoted X13 (or X31), X23 (or X32), and X33. (See Beck et al. [1] for more description of the numbering system, which uses 1 for the temperature boundary condition, 2 for the heat £ux, and 3 for convection.) The equations are presented in the form of algorithms that are readily programmed; a sample program for the X33 case is in the Appendix. These same eigenconditions are present in some spherical radial problems and also for problems with a thin high-conductivity surface ¢lm. Very large ranges of the Biot numbers are covered. Furthermore, this is the ¢rst article to present convenient equations for the X33 case, for two convective boundary conditions with different heat transfer coef¢cients on either side. When attempting to provide accurate zn for the X23 and X13 cases, it was noticed that zn is sensitive to the choice of end values. For this reason, it is necessary EIGENVALUES IN HEAT CONDUCTION 151 Figure 10. The deviation of computed b1 as a function of Bi1 and Bi 2 for the X33 case after two iterations. Figure 11. The deviation of computed b2 as a function of Bi1 and Bi 2 for the X33 case after two iterations. 152 A. HAJI-SHEIKH AND J. V. BECK Table 5. Calculated eigenvalue, zn , for RS13 and comparison with results after two and three iterations n Ba zn zn Zucher [2] Second iteration 1 ¡1.00 ¡0.95 ¡0.90 ¡0.80 ¡0.60 ¡0.40 ¡0.20 0.00 ¡1.00 ¡0.95 ¡0.90 ¡0.80 ¡0.60 ¡0.40 ¡0.20 0.00 ¡1.00 ¡0.95 ¡0.90 ¡0.80 ¡0.60 ¡0.40 ¡0.20 0.00 ¡1.00 ¡0.95 ¡0.90 ¡0.80 ¡0.60 ¡0.40 ¡0.20 0.00 ¡1.00 ¡0.95 ¡0.90 ¡0.80 ¡0.60 ¡0.40 ¡0.20 0:00 0.00000 0.39305 0.54397 0.75002 1.03511 1.24961 1.42428 1.57080 4.49362 4.50455 4.51552 4.53757 4.58181 4.62591 4.66952 4.71239 7.72518 7.73162 7.73807 7.75098 7.77685 7.80268 7.82842 7.85398 10.90408 10.90865 10.91323 10.92239 10.94073 10.95906 10.97735 10.99557 14.06617 14.06972 14.07327 14.08038 14.09459 14.10881 14.12300 14.13717 0.0000000 0.3853669 0.5422809 0.7593083 1.0527958 1.2644040 1.4320323 1.5707963 4.4934095 4.5045364 4.5156604 4.5378886 4.5821879 4.6261383 4.6695848 4.7123890 7.7252518 7.7317240 7.7381957 7.7511351 7.7769834 7.8027626 7.8284393 7.8539816 10.9041217 10.9087070 10.9132922 10.9224613 10.9407885 10.9590911 10.9773570 10.9955743 14.0661939 14.0697485 14.0733030 14.0804114 14.0946232 14.1088235 14.1230066 14.1371669 0 0.38537 0.54228 0.75931 1.05279 1.26440 1.43203 1.57080 4.49341 4.50454 4.51566 4.53789 4.58219 4.62614 4.66958 4.71239 7.72525 7.73172 7.73820 7.75114 7.77698 7.80276 7.82844 7.85398 10.90412 10.90871 10.91329 10.92246 10.94079 10.95909 10.97736 10.99557 14.06619 14.06975 14.07330 14.08041 14.09462 14.10882 14.12301 14.13717 0 0:385368098094658 0:542280885416155 0:759307689030632 1:05279429376987 1:26440357774200 1:432032236243418 1:570796326794897 4:493409457909064 4:504536385052075 4:515660437913874 4:537888582246557 4:582187926046559 4:626138290981610 4:669584780905963 4:712388980384690 7:725251836937707 7:731724026038532 7:738195664946898 7:751135101682629 7:776983426127267 7:802762575422832 7:828439313712818 7:853981633974483 10:90412165942890 10:90870704909461 10:91329224451455 10:92246127745309 10:94078851202827 10:95909110563603 10:97735698123371 10:99557428756428 14:06619391283147 14:06974851962217 14:07330303618783 14:08041143825164 14:09462320275459 14:10882348067746 14:12300659712559 14:13716694115407 2 3 4 5 a B ˆ hr =k ¡ 1. 0 to have highly accurate asymptotic values when developing zn for the X33 case. Also, for any given n, the zn for the X13 and X23 cases depend on one Biot number, whereas zn for the X33 case depends on two Biot numbers. This makes obtaining a simple relation for X33 more demanding. However, for less critical applications, one may use Eqs. (2a)^(2c) for X23, Eqs. (11a)^(11c) for X13, and Eq. (19) with E23 ˆ px23 =4 for X33. EIGENVALUES IN HEAT CONDUCTION 153 When using the algorithms in Table 1, the numerical accuracy for the X13 and X23 cases exceeds the self-imposed target of accuracy, that is, an error of less than 10¡7 . Also, this target of accuracy for X33 is satis¢ed when n > 1 and marginally satis¢ed when n ˆ 1. Therefore, the explicit relations that include one high-order iteration provide accuracy suf¢cient for all practical applications. The procedure to compute these eigenvalues for Cartesian coordinates is summarized in Table 1. The equation derived to obtain approximate eigenvalues in Cartesian coordinates applies equally to spherical coordinates; only minor modi¢cation is needed to accomplish this task. The procedure described here applies to spherical coordinates once gn;13 in Eq. (13c) is replaced by gn;13 ˆ 1 ‡ p 1 ¡ 0:8 2:18835 exp ¡dn2 Bi ‡ n x13 …1 ¡ x13 † ‡ n p …23a† and Eq. (23a) is modi¢ed for inclusion in the X33 case as gn;13 ˆ 1 ‡ 1 ¡ 0:8 2:18835 Bi1 Bi2 exp ¡dn2 ‡n x13 …1 ¡ x13 † ‡ n Bi1 ‡ Bi2 p …23b† These changes are needed because the Bi term in Eqs. (10) and (23a) may be replaced by a constant B that can go from ¡1 to in¢nity (see R03 in [1], p. 468); at B ˆ ¡1, dbn =dB ! 1. For example, using a solid sphere with radius ro , this constant is B ˆ hro =k ¡ 1. When Bi µ 0, the computed eigenvalues in Table 5 are compared with the tabulated data in [2]. The largest error is in the vicinity of Bi ˆ ¡1. Also, Eq. (23a) can be used successfully when Bi > 0; however, the data accuracy using Eq. (13a) is slightly less than those from Eq. (13c). For a sphere with inner radius ri and outer radius ro , the parameters Bi1 and Bi2 in Eqs. (16) and (23b) may be replaced by B1 and B2 , respectively. For the RS33 case described in [1], when the heat transfer coef¢cients at the inner and outer surfaces are h1 and h2 , one obtains B1 ˆ …h1 a=k ‡ 1†…b=a ¡ 1† and B1 ˆ …h2 b=k ‡ 1†…1 ¡ a=b†. All computations reported here were performed using Mathematica [10], and a sample program for X33 is in the Appendix. R EFER ENCES 1. J. V. Beck, K. D. Cole, A. Haji-Sheikh, and B. Litkouhi, Heat Conduction Using Green’s Functions, Hemisphere, Washington, DC, 1992. 2. Ruth Zucker, Elementary Transcendental Functions, in M. Abramowitz and I. Stegun (eds.), Handbook of Mathematical Functions, National Bureau of Standards AMS 55, p. 224, June 1964. 3. M. M. Yovanovich, Simple Explicit Expressions for Calculation of the Heisler-Grober Charts, 1996 National Heat Transfer Conf., Houston, Texas, Paper AIAA-96-3968, 1996. 4. R. A. Leathers and N. J. McCormick, Closed-Form Solutions for Transcendental Equations of Heat Transfer, ASME J. Heat Transfer, vol. 118, pp. 970^973 , 1996. 5. W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in Fortran. The Art of Scienti¢c Computing, Cambridge University Press, Cambridge, UK, 1994. 6. C. J. F. Ridders, IEEE Trans. Circuits Systems, vol. CAS-26, pp. 979^980, 1979. 154 A. HAJI-SHEIKH AND J. V. BECK 7. R. P. Brent, Algorithms for Minimization without Derivatives, chaps. 3, 4, Prentice-Hall, Englewood Cliffs, NJ, 1973. 8. C. Aviles-Ramos, K. T. Harris, and A. Haji-Sheikh, A Hybrid Root Finder, Proc. Fifth Integral Methods in Science and Engineering, Houghton, MI, 1998. 9. J. W. Stevens and R. Luck, Explicit Approximations for All Eigenvalues of 1-D Transient Heat Conduction Equations, Heat Transfer Eng., vol. 20, no. 2, pp. 35^41, 1999. 10. S. Wolfram, The Mathematica, 3d ed., Cambridge University Press, Cambridge, UK, 1996. 11. T. R. McCalla, Introduction to Numerical Methods and Fortran Programming, Wiley, New York, 1967. AP PEND IX This Appendix is concerned with using a simple high-order Newton method for ¢nding the root of the equation F …x† ˆ 0 …A:1† It is possible to use the ¢rst- or higher-order Newton method to ¢nd the roots of this equation. A ¢rst-order Newton method was not suf¢ciently accurate for a one-term approximate explicit equation. The second-order Newton with the square-root term because unstable when the Biot number was extremely large. This modi¢ed Newton method is used because it has nearly the accuracy of the second-order Newton method and it is numerically stable similar to the ¢rst-order Newton. Moreover, it can be applied to the X33 case, where it is desirable to accommodate a third-degree polynomial. Although this method is different from Bailey’s iterative method [11], they are conceptually similar. If x0 is an estimate of the root, h ˆ x ¡ x0 , then, using the Taylor series for F …x† ˆ F ‰x0 ‡ …x ¡ x0 †Š ˆ F …x ‡ h†, one can write F …x† ˆ F …x0 † ‡ F 0 …x0 †h ‡ M ˆ mˆ0 1 00 1 F …x0 †h2 ‡ F 000 …x0 †h3 ‡ ¢ ¢ ¢ 2! 3! Am hm ˆ 0 …A:2† where Am ˆ F 0 …x†=m! computed at x ˆ x0 . When h is small, the ¢rst-order Newton equation uses the ¢rst two terms in the above series to approximate h as h0 ˆ ¡A0 A1 …A:3† However, for a better approximation, without signi¢cant additional efforts, one can write h ˆ h0 ‡ Dh, where Dh½h if h is already small. Substituting for h in the polynomial represented by Eq. (A.2) and deleting the terms that contain Dh with a power of 2 or higher leads to a relation that determines Dh, Dh ˆ ¡ ˆ M m mˆ2 Am h M m¡1 mˆ1 mA m h M m m mˆ2 …¡1† Am …A0 =A1 † M m m¡1 mˆ1 m…¡1† Am …A0 =A1 † …A:4† EIGENVALUES IN HEAT CONDUCTION 155 When M ˆ 1, Eq. (A.4) yields Dh ˆ 0, which corresponds to a ¢rst-order Newton method. When M ˆ 2 it approximates a second-order Newton method, while M ˆ 3 improves the accuracy; however, it is rarely necessary to use M > 3. Example. As an example, consideration is given to ¢nding the roots of the equation F …x† ˆ x sin x ¡ B cos x ˆ 0 …A:5† which describes the eigenvalues for the X23 case. If x0 is an estimate of the root, then A0 ˆ F …x0 † ˆ x0 sin x0 ¡ B cos x ˆ 0 A1 ˆ F 0…x0 † ˆ …1 ‡ B† sin x0 ‡ x cos x F 00 …x0 † …1 ‡ B† cos x0 ‡ cos x0 ¡ x0 sin x0 A2 ˆ ˆ 2! 2! ˆ ‰…B cos x0 ¡ x0 sin x0 † ‡ 2 cos x0 Š=2 ^ cos x0 …A:5† …A:5† …A:5† Using M ˆ 2, the value of x is x ^ x0 ¡ A0 …A2 =A1 †…A0 =A1 † 1‡ A1 1 ¡ 2…A2 =A1 †…A0 =A1 † …A:5† The following symbolic function in Mathematica demonstrates the code for X33 when M ˆ 3. The ¢rst two arguments of this function are Bi1 and Bi2 , while the third argument refers to the nth eigenvalue to be computed. This function accepts all positive numbers, including 0 and 1, for either Bi1 or Bi2 . (*Mathematica function begins here and ends after the return statement*) eigen[bi1^ ,bi2^ ,n^ ]: ˆ ( cn ˆ (n¡3/4)*Pi; dn ˆ (n¡1/2)*Pi; d ˆ 1.22*(n¡1) ‡ 0.76; p23 ˆ 1/n; term ˆ Limit[Limit[(Sqrt[(b1 ‡ b2 ‡ b1*b2 ‡ cn¡Pi/4)/(b1 ‡ b2 ‡ b1*b2 ‡ d)]¡ (b1 ‡ b2 ‡ b1*b2 ‡ Sqrt[d*(cn¡Pi/4)])/(b1 ‡ b2 ‡ b1*b2 ‡ d)),b1! bi1], b2!bi2]; g23 ˆ 1¡1.04*term; x23 ˆ Limit[Limit[(b1 ‡ b2¡cn)/(b1 ‡ b2 ‡ cn),b1!bi1],b2! bi2]; x13 ˆ Limit[Limit[b1*b2/(b1*b2 ‡ 0.2 ‡ (b1 ‡ b2)*(Pi*Pi*(n ¡0.5)/2)),b1 ! bi1],b2! bi2]; g13 ˆ 1 ‡ x13*(1¡x13)(1¡0.85/n¡(0.6¡0.71/n)*(x13 ‡ 1)*(x13¡0.6¡0.25/n)); e23 ˆ g23*(p23*x23 ‡ (1¡p23)*Tanh[x23]/Tanh[1]); e13 ˆ 2*g13*x13; y23 ˆ cn ‡ Pi*e23/4; y13 ˆ dn ‡ Pi*e13/4; param ˆ Limit[Limit[b1*b2/(b1*b2 ‡ dnê2),b1! bi1],b2! bi2]; z0 ˆ (y13*param ‡ y23*(1¡param)); a0 ˆ Limit[Limit[((z0^ 2¡b1*b2)*Sin[z0] ¡(b1 ‡ b2)*z0*Cos[z0])/(b1 ‡ b2),b1! bi1],b2! bi2]; a1 ˆ Limit[Limit[((z0^ 2¡(b1 ‡ b2)¡b1*b2)*Cos[z0] ‡ (2 ‡ b1 ‡ b2)*z0*Sin[z0])/(b1 ‡ b2), b1! bi1],b2! bi2]; a2 ˆ Limit[Limit[((1 ‡ b1 ‡ b2)*Sin[z0] ‡ 2*z0*Cos[z0])/(b1+b2),b1! bi1], b2! bi2]-a0/2; 156 A. HAJI-SHEIKH AND J. V. BECK a3=Limit[Limit[(((6+3*(b1+b2) +b1* b2-z0*z0)*Cos[z0]-(6+b1+b2)*z0*Sin[z0])/6) (b1+b2),b1! bi1],b2! bi2]; e=-a0/a1-(-a3*a ^ 3+a2*a0 ^ 2*a1)/(3*a3*a0*a0*a1-2*a2*a0*a1 ^ 2+a1 ^ 4); z1=z0+e; Return[z1] (*Calling the function to calculate eignevalues with 10 digit output*) N[eigen[1,10,1],10] Out[2]=1.87530781 N[eigen[1,01],10] Out [4]=0.86033358 9 N[eigen[In¢nity,1,1],10] Out[6]=2.028757838