Further Approximations for Elliptic Integrals By Yudell L. Luke* Abstract. The present paper develops approximations for the three kinds of elliptic integrals based on the Padé approximations for the square root. The work includes and extends our previous work on the subject to provide efficient approximations over a larger part of the complex k and <¡>planes. 1. Introduction. In previous work [1], [2], we gave some approximations for the three kinds of complete and incomplete elliptic integrals which were derived by using Padé approximations for the square root. The expansions are valid in a large sector of the complex plane and are far more powerful than the analogous representations based on the binomial expansion for the square root. Nonetheless, the convergence properties of our previous developments weaken for k2 sin2 </>near unity. In this paper, we develop new representations like those of [1], [2] to correct this deficiency. The present study is suggested by analogous work for power series developments of the first two kinds of elliptic integrals. In this connection, see the recent study by Van de Vel [3] and the references quoted there. For convenience and completeness, we also give the representations previously found. Our proofs are very sketchy as the details are akin to those previously used. 2. Approximations for the Square Root and Other Data. (1) (1 - z)-1'2 = (2n + I)"1 ¿ (2) 6,(1 - z sin2 ft.)"1 + VnOz) , (1 - z)1'2 = 1 - 2z{2n + I)"1 ¿ sin'°", 77.-1 1—2 (3) em = 1 if m = 0 , ím = 2 (4) 6m = m7r/(2ft + 1) . + WAz) , COS 6n if m ^ 1 , Let t _ 2- (5) 2 ±2(1 - z)112 where the sign is chosen so that \el\ lies outside the unit disc which is possible for all z except z ^ 1. Then 4e-(2n+3/2)f (6) VAz) = —2 z1/2(l - e-f)[l + e -(2n+l)¡--| ' Received July 30, 1969. AMS Subject Classifications. Primary 33.19, 41.17. Key Words and Phrases. Elliptic integrals, approximations by rational functions. * This research was supported by the U. S. Atomic Energy Commission under Contract AT(11-1)1619. 191 License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use No. 192 YTJDELL L. LUKE ¿<2(l (7) Wniz) = _ e-r)e-(2«+i/2)f [1 _ e-C2»+l)f] and for z fixed, (8) lim \Vn(z) and Wniz)] = 0 , 2 ^ 1, |arg (1 - z)[ < v . n—*oo It is also convenient to have representations like (1) and (2) with z replaced by -z. Thus (9) (1 + z)-1'2 = (2n + I)"1 ¿ 6m(l + z sin2 6A~l + Rniz) , 771=0 1 V-» (10) (1 + z)1'2 = 1 + 2z(2n + l)-1 ¿\ -2- Sin Ott. i 1 + z cos 6m i C7 / \ + Sniz) For the error terms, let 1/2 , = 2 + z±2(l-rz) (11) 2 where the sign is chosen so that \ev\ lies outside the unit disc, which is possible for all 2 except —1 2: z ^ — co. Then (12) Rniz) = (13) <S„(z) Ae-(2n+3/2)7) 77 2,/2(l + e-')[l ZW»(1 + - e-<2"+"'] ' e-,)e-(2n+l/2), (2n+l)7,, [1 + c-l«H-i»] ' and for z fixed, (14) lim {Rniz) and Sniz)} = 0, z j¿ -1, |arg (1 + z)| < *-. n—»00 Finally, we have need for some integrals: (15) Hid,a2)= f--f^l"J«l-o 2s (16) -1 H (6, a ) = c arc tan (c tan 0) , |arg (1 — a )| < ir, (17) H ((9,a2) = (2g)-1 In H(6,a) (18) a c = (1 — a 24,1/2 ) |arg (1 ± ¿ctan0)| 1 4- grtan 0 1 — ¡7tan 0 < ir, g = (a — 1 ) ¿ = (— 1) 1/2 , g tan 0 real . = tan0 if a = 1 , = ir/2e if 0 = tt/2 and |arg (1 — a )\ < it , = 0 (19) sin if 0 = ir/2 and a2 > 1 . / -Sm22 " , da = a"2[ff (0, a2) - 0] . •'o 1 — a sin a In the above, the integrals are to be interpreted in the Cauchy sense when a2 sin2 0 License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use FURTHER > 1. Indeed, appropriate. /oa\ (20) ría APPROXIMATIONS throughout 2s this paper, f" FOR ELLIPTIC we interpret cos ada [" 2s ,„4-i, = (2c) 1 + In-^— c sin 0 1 — c sin 0 (21) , integrals ■2 . ((1 1 — - a« )) sin a c = (1 - a2)1'2, | arg (1 — a2) | < ir , c sin 0+1 c sin 0 — 1| (22) J(6, a2) = (2c)_1 In (23) J(d, a2) = sin0 (24) J(0, a2) = g~l arc tan (¡5sin 0) , 3. Approximations in this sense when C°S ada Ji6, O) = J -2 , 2 ■ 2 = /-2s J o COS cos a + a sin a 'i 1na J(6,a) 193 INTEGRALS csin0 j¿ 1 | arg (1 ± c sin 0) | < it. c sin 0 real if a2 = 1 . g = (a2 — 1)1/2, g sin 0 real. for the Elliptic Integrals of the First and Third Kinds. Let Fi4>, k, v) = / (1 - v2 sin2 a)-1 il - A sin2 a)-1/2o (25) v sin <j>9e 1 , |arg (1 — fc )| < ir , |arg (1 — k sin2 <£)| < tt . Using (1) and (15) — (18), we get n+,k,,) -f£$ %Ál - 5 sin2öm)_I 2/c (26) Qni<t>,k,v) = Qn(<¡4, fc,P) = / •A4sin 6mHj4>, k sin 0m) (2ft + l)v*¿A 1 - ik2/v2) sin2 0„ (1 — i-2 sin2 a) ^„(/c2 sin2 a)da 2c-(2"+1)r tan <b . :^ + 0(ft-2)] + 0(e-4nr) (1 - *<2sin2 <p)(4ft + 3) L~ ' 4ft 4- 3 Xi = 1 cosh f — 1 2iv sin 0 (cosh f — 1) sinh f cos <j> (1 — v sin 0) sinh f 0 ^ an odd multiple of (27) 0.(t'*'')-[ 2tt sinh f (4ft + 3) (cosh f + 0(e-ter) " 1/2 _(2„+i)f r , _ (2 - fc2)(cosh r + 1) 1 y2(coshr-l) Ssinhr " 2 + (i_,2)sinhf: where ef is defined by (5) with z = k2 sin2 <j>. Clearly for fixed </>,k and ?, under the conditions given in (25), License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 2 ' 194 YUDELL L. LUKE (28) lim Q„(0, k,v) - 0. In (21), the coefficient of Gi<t>,v2) can be replaced by (1 - k2/v2)~V2 - F„(fc2/V2) provided |arg (1 — k2/v2)\ < x. This should be used only when the magnitude of V„ik2/v2) does not exceed the magnitude of QA<i>,k, v), which happens approximately when \es\ with z = k2 sin2 0 exceeds |ef| with z = k2/v2. In the case that 0, k, and v are real, this happens approximately when 1 ;£ v2 sin2 0. Some numerics illustrating these approximations and the striking realism of the error formulations are given in my earlier works. It is clear that the accuracy weakens for values of k2 sin2 <j>near unity. To correct this deficiency, we proceed as follows. We can write „, . s / b (0, k, V) - J ..„„s secado; y 2 • 2 w, o (l — j7 sin k'2 = 1 - A , . ,,2 , a)(l + k tan 2~TÏ72 » a) |arg (1 + k'2 tan2 0)| < t . Now use (9) and (20) - (24). Thus Fi<b, k,v) =-l--r ¿-<Al-k'2sm2em) (2ft + 1)(1 - v2) 77-ti l - 2 V (fc'2/(l ,2 - Ji<t>,1 - «') Ta (2ft + 1)(1 - y2) ' ST' ä „2)) sin2 em 'Si- (fc'7(i _ „»)) sin2 + Sn{<t>,k,v), tan , , /" sec sec aRnik ,k,v) -- ali S„(4>, k,v) = J/-——■'o ° 11 — v sm 70Q4. 2e~(2n+1)vsin 0 &(*. *,-)-. (31) a)da a (1 I p "", ■ 2 "7" QsF1 - 4ft7^5 + 0(n-2 — j/ sm 0) (4ft + 3) L -f- «3 sin2 0 sinh g f 2(1 ~ y2) "=1+cosh,-l Ll-^sin2^1]- il + 0(e-4""), sinh 77 cosh where e" is defined by (11) with 2 = fc'2tan2 <j>.Then under the same conditions as in (29), with <b,k and v fixed, (32) lim Sni<t>, k, v) = 0 . 71—»0O The approximation (30) is convenient even when fc'2 tan2 0 > 1. Notice that if 0 -> tt/2 and fc -> 1 so that cos2 0 = fc'2, then fc'2tan2 0 -^ 1 and e"' -^3 -23/2 =0.172. The coefficient of ■/(</>, 1 — v2) in (30) can be replaced by -,2[(1 - ,2)(fc2 - V2)]-1'2 + V2(l - V2A'Vnik'2/il - V2)) provided [arg (1 — fc'2/(l — v2))\ < -k. However, this should not be introduced unless the magnitude of F„(fc'2/(1 — v2)) is approximately equal to or less than the magnitude of Sni<l>,fc, v). To get a useful approximation when 0 is near 7r/2, we write License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use FURTHER APPROXIMATIONS (T/2)-« F\^,k,Aj-F{4>,k,v) = /o' 195 FOR ELLIPTIC INTEGRALS Sec ada a)(fc' + tan 2 a)\l/2 (1 - ¡/cos tan 6) ' de f* sec 0(1 + fc'21! K ' i - v2+f- fc' tan (33) fc' tan i/- = cot 0 , y2sin 0 ^ 1 , [arg (1 + fc' tan ^)| < it . Now use (10). We find that Ki'/C) ") " F(*'fc'^ = (1 ~ v2)'1JV'(T~7)) (1 2s X 1(34) 71 • 2(1 - v2) '_) \p_sm 2ft + 1 2 n 0m_ 777=11 — (1 — j,2) cos2 6mA • 2 a Bill sin 0„ 4r-4 + 2ft + T¿íi - (1 - v2) cos2 6„ (1 — X /(-A, k'2 cos2 0m) + i¥n(0, fc, v) , Mn((f>,k,v) = Mni<t>,k,v) (35) f* sec eSnjk'2 tan 6)d8 Jo 1 - v2 + fc'2tan20 2(cos 0/(1 + sin 0)) + (1 - sin 0) (1 - fc sin 0r"(l X 1+ 4n + 1 <7= 2 sin 0—1 -/sin 0)(4ft+ + 0(n2) fc' sin 0 1 — fc sin 0 1) / COS0Y U\l + sin 0/ ' 2(1 — v ) sin 0 1 — v~sin2 0 In (34), the coefficient of J(ip, fc'2/(l — v2)) can be replaced by v(l — v2)*1— (1 — v2)~lWn(l — v2) provided |arg v\ < ir/2. This replacement should not be used unless approximately the magnitude of Wn(l — v2) is less than the magnitude of Mn(4>, fc, v). If v and 0 are real, e_f = (1 — v)/(l + v) when 2 = 1 — v2. Thus, use the alternative form provided (1 - »0/(1 + v) t% (1 - sin 0)/(l + sin 0) , i.e., when sin 0 ^ v < 1. Under the same restrictions as in (33), with 0, fc, and v fixed, (36) lim Mni<t>, fc,v) = 0 . Our technique is applicable to derive an expansion relating K(k) = F{w/2, fc, 0) and K(k'), where fc'2 = 1 — fc2,since it can be shown that (37) K(k) + w-*Qnk'2)Kik') = - -| /"* -&JË 7T •'O sin2 0)(i0 (1 — fc' sin22axl/2 0) License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 196 YUDELL L. LUKE Thus = 2(2« + I)"1 In 2 + 2 £ a„ K(k) + n'1 (hi k'2)K(k') L (38) am = (1 - An(k) = -— (39) + -4„(fc), In (1 + a,„) 774=1 A sin2 0m)1/2 , / (lnsin20)r,i(fc'2sin20)d0, ir •' u ^'^'VKf-ini-äMt^ii+oc.-.,] + 0 Hence with fcfixed, |arg fc| < 7r/2, (40) lim4„(fc) = 0. 71—700 4. Approximations for the Elliptic Integral of the Second Kind. Let E(4>, k)= f (1 — fc sin a) da •7 n (41) |arg (1 — fc )| < ir, |arg (1 — fc sin 0)| < n . Here we use (2) and (15) — (19). Thus n E(4>, fc) = (2ft + 1)0 - 2(2n + I)"1 £ tan2 0^(0, (42) m=1 rn(0, fc) = / Jo T U IA Wn(k2 sin2 a)da . 2e-(2"+1Ktan0(l-fc2sin20) + 0(e-ini) - -• \ g. . "1 , 2fc2sin2 0(cosh f + 1) 0"! = fc2cos2 6m) + Tn(4>,k ), , . sinh f (cosh f - 1) -Scos 0 sinh f ' 0 t^ an odd multiple of ~¡r (43) 77f71" i\ _ f 2in;coshr - 1) T7'-cm-ik,. H 2 ' V ~ ~L (4n+1) sinhfj (J2 6 (1 (9fc2sin2 0 - 2) (cosh f - 1) 1 8 sinh f 2 ' k sin 0) License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use FURTHER APPROXIMATIONS FOR ELLIPTIC 197 INTEGRALS where ef is defined by (5) with z = fc2sin2 0. Under the same conditions as for (41) with 0 and fcfixed (44) lim Tn(4>,fc) = 0. 7i—7(70 Numerical examples illustrating (42) and the remarkable formulas for T„(0, fc) are given in [1], [2]. After the manner of getting (30), we find ,-,,, ,s 7, s . , . 2fc'2 efficiency of the error A sin2 0mJ(0, fc'2 cos2 8A) i?(0, fc) = p„ik) sin 0 + r—- 2,-—2- ¿n -+■i m=i 1 — fc' cos 0m + Uni<t>,fc), |arg (1 + fc'2tan2 0)| < w , (45) ,,, ^k) , 2fc'2 A sin20m = l-2n-+-l£il- f*cos aS„ (fc'2tan2 a)da ?7„(0, fc) = / Jo rr /_. ;\ (46) , , 2ku (l - fcVn+1 fc'2cos27„ = k + Y^u' > . 2e""<2n+I),(l — fc2SÍn20) SÍn0f (4ft+ 1)-L1+iftTT 2 sinh 7? cosh 1?+ 1 U=\T+k) Ji_ _. _2s . „7 _4n,x + 0(n )J + 0(e }' 3 (cosh 77+ 1) cos 0 sinh 7? ' where e" is defined by (11) with 2 = fc'2tan2 0. Thus for 0 and fc fixed and restricted as in (45), (47) lim U«i4>,k)= 0. n—700 In (45), we can take p„(fc) = fc provided that |2fcw/(l — w)| < |í/n(0, fc)|. In this connection we notice that if fc and 0 are real, then (1 — fc)/(l + fc) < e-' when fctan 0 > 1. A straightforward (48) analysis shows that E{k) - E(<t>,fc) = fc'2F(o;,fc,fc) , fc'tan w = cot 0 , and so we can apply the expansions previously developed as appropriate. A relation analogous to (37) for the complete elliptic integral of the second kind is (49) E(k) = ^E(k') + ^K(k') - ^ {2 + In fc'2| f/2 ,. - J° t'"!^ (l-fc'sin<0) 2 _ 2/V2 r/2sin20(lnsin20)4J0 T K (1 -fc'2sin20)1/2 and from this one can derive the analogs of (38) — (40). The result would be rather complicated and not as efficient as the known expansion for E(k) in powers of fc'2. Rather than use the expansion which could be found from (49), a much more efficient procedure would be to employ the Legendre relation, License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 198 (50) YUDELL L. LUKE E(k)K(k') + Kik)Eik') - K(fc)K(fc') = x/2 , and the appropriate expansions for K{k) and K{k'). Midwest Research Institute Kansas City, Missouri 64110 1. Yudell L. Luke, "Approximations for elliptic integrals," Math. Comp., v. 22, 1968, pp. 627-634. MR 37 #2412. 2. Yudell L. Luke, The Special Functions and Their Approximations, Vol. 2, Academic Press, New York, 1969, pp. 269-273. 3. H. van de Vel, "On the series expansion method for computing first and second kinds," Math Comp., v. 23, 1969, pp. 61-70. License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use elliptic integrals of the