Appendix F Asymptotic Expansion of Single Integrals
Asymptotic analysis is a powerful analytical approach to obtaining elegantly simple analytic approximations to problems that contain either a parameter or a variable whose magnitude becomes either large or small in comparison to some value that is characteristic of the problem. Its elegance lies in the fact that the results may usually be expressed in a single dominant term that contains all of the essential physics of the problem instead of through the subtle interaction of a large (perhaps infinite) number of terms in a summation. The basic idea behind this general approximating method of analysis may be illustrated by the evaluation of the real exponential integral [1] Z 1 1 t E1.x/ e dt; (F.1) x t where x>0is real-valued. This function possesses the convergent series expan- sion [2] X1 xn E .x/ D C ln.x/ C (F.2) 1 nnŠ nD1 P 1 1 for x>0, where limn!n kD1 k ln.n C 1/ D 0:57721 : : : is Euler’s constant. Although this series converges for all positive values of x, it becomes computationally useless for x 1. To obtain a useful expression for the value of this function for large values of its argument, repeated integration by parts results in [1] ˇ Z t ˇ1 1 e ˇ 1 t E1.x/ D ˇ e dt t t 2 x ˇ x Z x t ˇ1 1 e e ˇ 1 t D C ˇ C 2 e dt x t 2 t 3 x x e x 1Š 2Š 3Š nŠ D D 1 C C C. 1/n C R .x/; x x x2 x3 xn nC1 (F.3)
777 778 F Asymptotic Expansion of Single Integrals where the remainder after n terms is given by Z 1 e t R .x/ D . 1/nC1.n C 1/Š dt: nC1 nC2 (F.4) x t
Since the integral appearing in this remainder term is bounded by e x=xnC2 for x>0, then the magnitude of this remainder term is bounded as
.n C 1/Š jR .x/j < e x: (F.5) nC1 xnC2 The remainder after n terms is then seen to be bounded in magnitude by the magni- tude of the first term neglected in the series summation e x 1Š 2Š 3Š nŠ S .x/ D 1 C C C. 1/n : (F.6) n x x x2 x3 xn
However, if one considers the expansion given in (F.6) as an infinite series, the result is divergent. Nevertheless, for sufficiently large values of x>0, the series summation given in (F.6) is rapidly convergent for a finite number of terms n.An estimate of the optimum number of terms to be used in this expansion for a given value of x may be obtained from the ratio of successive terms as jun.x/=unC1.x/j D n=x 1, so that the optimum number of terms to be used in the summation Sn.x/ for an estimate of E1.x/ for a given large value of x is approximately given by the greatest integer in x, as illustrated in Fig. F.1 when x D 5:7. In that case the 5 optimum number of terms is given by n D 5 where 5e Sn.5/ D 0:8704, which is in 5 good agreement with the actual value of 5e E1.5/ D 0:8663. Inclusion of additional terms in the summation only results in a decrease in accuracy. Most importantly, since the remainder after the first (or dominant) term becomes exponentially small as x !1, the approximation E1.x/ S1.x/ becomes increasingly accurate as x increases. This example then leads to the following distinction between an asymptotic expansion and a power series expansion of some function: For the power series expansion XN f.x/Š un.x/ nD0 of a given function f.x/, the approximation to the value of f at some fixed value of x improves in some well-defined sense as N !1, while for an asymptotic expansion
f.x/D Sn.x/ C RnC1.x/ of that function, the approximation to f.x/by the series summation Sn.x/ improves in some (as yet undefined) sense for fixed n as x !1. The first (or dominant) term F Asymptotic Expansion of Single Integrals 779
1
0.95
0.9 (x) n
S 0.85 x xe
0.8
0.75
0.7 2 3 4 5 6 7 8 9 10 n
Fig. F.1 Dependence of the series summation Sn.x/ approximation of the exponential integral E1.x/ for x D 5:7 as a function of the number n of terms in the summation. The open circles x connected by the solid line segments describe the values of the quantity xe Sn.x/,whilethedashed curves describe upper and lower envelopes to these values. Notice that the approximate values 5 oscillate about the actual value of 5e E1.5/ D 0:8663
Fig. F.2 Statue of the Norwegian mathematician Niels Henrik Abel (1802–1829) wrestling with a sea serpent on the royal palace grounds in Oslo, Norway. (Photograph by K. E. Oughstun) in the asymptotic expansion of of f.x/ represents the asymptotic approximation of that function as x !1. However, care must always be taken to ensure that the given asymptotic expansion is not only well defined mathematically but is also properly applied and interpreted. If not, critical errors may result. Such was the motivation for Abel (see Fig. F.2) to lament in 1828 that 780 F Asymptotic Expansion of Single Integrals
Divergent series are the invention of the devil, and it is a shame to base on them any demon- stration whatsoever. By using them, one may draw any conclusion he pleases and that is why these series have produced so many fallacies and so many paradoxes. . . A brief outline of the essential theory is presented here. A more detailed de- velopment may be found in the texts by E. T. Copson [3], N. Bleistein and R. Handelsman [4], and J. D. Murray [5].
F.1 Foundations
Definition 1. Spherical Neighborhood. A spherical neighborhood of a point z0 is given by the set of points in the open disc
jz z0j <ı (F.7) if z0 is at a finite distance from the origin (i.e., if jz0j < 1), while it is given by the open region jzj >ı (F.8) if z0 is the point at infinity. Definition 2. O-order. Let f.z/ and g.z/ be two functions of the complex variable z that possess definite limits as z ! z0 in some domain D. Then
f.z/ D O.g.z// as z ! z0 (F.9) if there exist positive constants K and ı such that jf.z/j Kjg.z/j whenever 0< jz z0j <ı.Ifjf.z/j Kjg.z/j for all z 2 D, then f.z/ D O.g.z// in D. Definition 3. o-order. Let f.z/ and g.z/ be two functions of the complex variable z that possess definite limits as z ! z0 in some domain D. Then
f.z/ D o.g.z// as z ! z0 (F.10) if there exists a positive constant ı such that jf.z/j jg.z/j for any >0when- ever 0