Beruflich Dokumente
Kultur Dokumente
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The Royal Society is collaborating with JSTOR to digitize, preserve and extend access to Proceedings:
Mathematical and Physical Sciences.
http://www.jstor.org
Formulas for higher derivatives of the Riemann zeta-function are developed from
Ramanujan's theory of the 'constant' of series. By using the Euler-Maclaurin
summation methods, formulas for C(n)(s), (n)(1 -s) and C(n)(O) are obtained.
Additional formulas involving the Stieltjes constants are also derived. Analytical
expression for error bounds is given in each case. The formulas permit accurate
derivative evaluation and the error bounds are shown to be realistic. A table of '(s)
is presented to 20 significant figures for s = -20(0.1)20. For rational arguments,
(l/k), '(1/k) are given for k=-10(1)10. The first ten zeros of '(s) are also
tabulated. Because the Stieltjes constants appear in many formulas, the constants
were evaluated freshly for this work. Formulas for the yn are derived with new error
bounds, and a tabulation of the constants is given from n = 0 to 100.
1. Introduction
The Riemann zeta-function is defined by
(1)
(s) = En0 (Res >). (1)
n=1 n
The series on the right is absolutely convergent for Res > 1 and divergent for
Res < 1. The Riemann zeta-function has a simple pole at s = 1. There is no
singularity other than the pole at s = 1. For Res < 1, c(s) is defined by analytical
continuation through the functional equation
(l - s) = 2(27)-S (s8) (s) cos (%is). (2)
In the negative-half plane c(s) and its derivatives are oscillatory (see figures 1 and 2).
By taking the logarithmic derivative of (2) one obtains
'( -8s)/S(1 -s) = ln 2 + (lCs)-f(s)-'(s)/
tan (8).
where An= (-l)nyn/n! are the Stieltjes constants. The zeta-function has many
integral representations. One such formula that will be needed later is
m-~l t)
nl-S(
B2k 8s-k2
n-s-2k+l 8 In n- - 2-2
n), (7)
l) (sk)
k-i + 1slnn+-1- +R(s,
where Rm(s,n) has a form similar to (6b). The summation term inside the curly
brackets drops out when k = 1.
The series involving the Bernoulli terms in (6a) or (7) are divergent for all n. Here
the sum of such asymptotic series is obtained in the Poincare sense. Since f(2m)(x) and
f(2m+2)(x) are of the same sign in (n, oo), the remainder is
The constants An may be calculated with the help of a table of y. found at the end
of this paper. A method for estimating the truncation error will be found in ?5 (a).
The constant may also be evaluated from the Abel-Plana form of the Euler-
Maclaurin sum formula by complex integration (Hardy 1949; Evgrafov 1961;
Olver 1974). Ramanujant gives (5a) without the error term for his definition of
'the constant' of the function f(x). Hardy has removed some of the difficulties with
this definition (see Berndt's review (1985)), and remarks that the Euler-Maclaurin
constant provides another definition of the sum of divergent series. Hardy calls it the
Ramanujan sum because Ramanujan's work on divergent series was based mainly on
this definition. This sum may be quite different from the Abel or the Cesaro sums.
Ramanujan (1927) had communicated several examples in his letters to Hardy.
Consider now the function f(x) = xm, where m is an integer. Then
n- m
m=
(m+ )- (m 8-t) Snm-s+l (m >0). (9a)
j=l s=O
t Ramanujan writes this in his note-book (1957, vol. 1, p. 79): 'The constant of a series has some mysterious
connection with the given infinite series and it is like the centre of gravity of a body. Mysterious because we
may substitute it for the divergent infinite series.'
Proc. R. Soc. Lond. A (1995)
(nn
C(1) = lim { I- ln n}-= ;
n->oo ?=l 9
C(O)-() = iln27,
=--(--1)+1, C(-l)
C(-2) = -r(-2), C(-3) =- (-3)
and so on. When m is even, C(m) =(- ) 1)m 7(27r)-m-1T(m+ 1)(m+1). Substi-
tuting numerical values from table 4 we find agreement with Bendersky's calculation.
t Ramanujan gives many numerical examples. Berndt has examined carefully Ramanujan's methods
(Berndt (ed.) 1985). We take one example here. On p. 97 (Ramanujan 1957, vol. 1) one finds this entry: 'Show
that the constant in the series 10x/1 + 100\/2+ 100V3+ 1004 +... + 100?Vx
is -0.4909100'. Using a programmable
calculator Berndt finds
and comments that the alleged accuracy apparently cannot be obtained from the series. But Ramanujan's
method may have been like this. Since the constant is C =
-(- -i), by taking n = 10, m = 2 in (6a) one obtains
~100 = 100.57794416. Then from the functional relation one finds (--100o) =-0.49090995,
(lO) which supports
Ramanujan's statement.
Proc. R. Soc. Lond. A (1995)
- -(22n i F 2n
ns=-^ 1 1212
=V -2n+ \1Ztanl 2y n L(2n+ 1)ln ((2n+ 1)/27i)J'
n Sn c(S)
2 - 2.717 262 829 2045741016(0) 9.159 8901199034618401(- 3)
3 -4.936762108594947 8689(0) - 3.9864416636707504317(- 3)
4 - 7.074597145007145 7343(0) 4.1940019580456264741(- 3)
5 - 9.170493162 785 828005 4(0) - 7.850880657 6886855822(- 3)
6 - 1.124121232537534351 1( + 1) 2.2730748149745047523(-2)
7 -1.3295574569032520385(+ 1) -9.3717308522682935624( -2)
8 - 1.533872907364828182 1( + 1) 5.2058968223620912046(- 1)
9 - 1.7373883342909485265(+ 1) -3.7435668234818147277(0)
10 - 1.940313 325717 6569932(+ 1) 3.3808303595651664654(+ 1)
11 - 2.1427902249083563532(+ 1) - 3.744188518657 6224650(+ 2)
Note that C(-- 1) is the natural logarithm of the Glaisher constant in closed form
(Glaisher 1878).t In an analogous manner the constants of Sm = E kmIn F(k) may be
obtained because the series is reducible to sums involving E kmIn k.
and
Theorem 3. The 'constant' of the function f(x) = (lnn x)/x is C(s) = (-l) )(s),
Res>-1 -, s # 1.
Proof. This is a generalization of the preceding theorem. To prove this we first note
l- nn-i x rlnn x nn
x (A)
-nj X+i dx+sJ x++1dx (A)
n
Inn-l x
dx-s rlnn dx = Ilnnx - Inn x
dx. (B)
xS Xs xS_i Xs
-2 2q-2 n 11 (q - 41
I
=-2n+l- +22-2 22+2 (I- 2)
nfl f 7l n2n -T2
where p = In (n/n), q = arctan (2p/n). Here the principal value of the inverse function
will be taken. For n > 15 the following approximation is adequate,
2 (2 2n+ 1 1
s =-2n+1--arctan-In?
2n + +0 , n-?o.
n (15b)'
7r (2n l +l)ln ((2n + )/27c)
Proc. R. Soc. Lond. A (1995)
1)! n
f(r) = (1)r-n(r-
=(-i ]
S(-
nk-1
l)Pn-k-lk, r-1,
nk r> (6a)
(16a)
Xr k=O
f
) =
(-I)r()x-s-r, r > O, (16b)
where
4 (s+2 n n! Inn-k a
0 (n- k) i (s + 2m- 1)
(2ca)2m(+2m-1)-
+2m,-
n ~n 2 -
j ---om-l)1n (
2m-i
(2m)! ( -[n-1 j! lnW a
k=o (j-i) = -( i) ' (s + 2m - 1)
< 0 < 2. (18b)
Proc. R. Soc. Lond. A (1995)
Ilnr k r!
r (s-? '+k
__ _(IcYr+k(5 n. (19)
k=l
A;=l ^S (S8-
^ ^^)r+l =0
k=0 k! ( 9)
For the proof see Berndt (Ramanujan's Notebooks 1985, p. 224). In this paper we shall
restrict r to the non-negative integers. The yn are closely related to the Stieltjes
constants although it is not evident if Ramanujan knew of Stieltjes's work. (Hardy
(1912) also studies the constants without any mention of Stieltjes.) We now combine
(13) and (19). Since lim,, (lnax)/x/ = 0 for a, f any two positive numbers (Knopp
1964, p. 62) we obtain this alternative expression for \(n)(s):
( n+k
E ( 1)k
) )n (s-
(S ~)
+l ? =01)"(n,
kk=0 k (s- 1)k+Rm(s) (20)
Using Berndt's inequality lyk,/k! < 4/kn7 (Berndt 1972) we estimate the remainder
Rm(n, s) in the following way. Consider s = 2 and n = 4. Then
(- 1)kYC 4 +1)(k2)(k3)
IRm(4,2)1 (
ck=m
k=m k! 4c n k=m
,nk=m Ick
4 ? (m+j+l)(m+j+2)(m+j+3)
-m+4 E
7
j=0
Let Sn(z) = S=ojn/zj. Then for n some positive constant Sn(z) is convergent for
Izl> 1, and the sum is given by the recurrence relation Sn+(Z) = -z(d/dz)Sn(z),
n = 1,2, 3,... (Ferrar 1959, pp. 6-7, 172). Thus S0 = z(z- )-l, Sl = (z- 1)-2, etc.
Then
2)1
mIR(4, M+4 (m +l) (m+ 2)(m+3) [(m + l) (m+2)+(m + 2)(m + 3)
+ (m+3)
(m+ 3((n
] + 1) (m + 2) (7C2 + 4K+ 1)
(7r- 1)2 (IC- 1) ? (Ic- 1) f
Thus for the remainder IRm(4,2)1 to be less than 10-10 we take m > 27. In practice
m = 9 is adequate, in which case the neglected term in Iy131/9!- 7.6 x 10-11. This
suggests that the Berndt inequality could be improved.
We can improve the error estimate. Consider this inequality due to Israilov (1981):
The ck,r are given in (28) below. Israilov gives the first three C(k) as C(1) = ,
C(2) = 7, C(3) = 13. With some difficulty we find the next two: C(4) = 49, C(5) = 49.
If we take k = 5 and repeat the calculation we find that IRm(4,2)1 d 10-10 if m > 14.
Thus the Israilov inequality improves the error estimate. Note also from (20) that
(n)(2) (- l)nn !, n - oo. The first five derivatives of c(s) at s = 2 are given in table
2 below. Computation by formulas (18a), (20) agree when appropriate number of
terms indicated by the error analysis are taken.
r
k=O
We calculated the first five derivatives of 5(1-s) at s = 2 using both methods and
found complete agreement. See table 2 for the result.
In this paper (n(0) will be obtained directly from (12). The lower-order derivatives
are easy: -'(0) = ln2n, "(0) =-2+o(1), etc. In general,
a-i n-1 (- Ina)k
(- 1)n(n)= C(0) n(0) =E lnnk-(a-) ln a-(-)n a(n !) (
k=l k=0
P.a>l. (23a)
Proc. R. Soc. Lond. A (1995)
n ((-)
(n)(s) ( - S)
1 - 9.3754825431584375370( - 1) - 1.654211437004509292 1( -1)
2 1.9892802342989010234(0) -2.5020442410960038929(- 1)
3 - 6.000145 802 843044 865 6(0) - 3.825 315 249197 722 934 8(- 1)
4 2.400148639373646157 1(+1) -7.4748643283702206332(-1)
5 - 1.2000082433327181677(+2) - 1.8738512392950424338(0)
The Pm, are given in (17). Without proving first that f(r)(x) is of constant sign in
(a, oo), we may not apply the error bounds discussed in ?2. We give below an error
bound by considering a comparison function f()(x), which is monotonic in (a, oo) (cf.
(16a)):
-
I (r-
X) I r > 1.
f\r)(x)' nk r_l// lnkx,
k =O
Since /m n > 0 and x > 1, If(r)(x)l /()(x). From (5b) the remainder is
In arriving at the above result the order of the integration and summation was
interchanged (Knopp 1964, p. 351). A method that gives a sharper estimate is
discussed in ?6a. For some numerical example take n = 2, a = 20, m = 10, and let
"(0) be the computed value. The exact value of "(0) is known from a result due to
Ramanujan (see ?6b). The error estimate shows IR10(oo)l< 7.0x 10-24; the term
neglected is -2.7 x 10-25, whereas the actual error is 1I"(0)-"(0)1 = 2.5 x 10-25
The actual error is less than the error estimate. Note that Cn(O)- (- 1)+n !, so that
(n)(0)-->n! as n- oo. The first twenty (n)(0) calculated from (23a) are given in
table 5.
The (n)(0) may also be calculated by Ramanujan's method. From (20), which is
valid at s = 0, we have
m-1
(-1)(n() =--(--1)nn!+ E ++Rm(n, 0). (24)
k=O
The (n)(0) are readily calculated with the help of a table of the Stieltjes constants
(table 5). Apostol (1985) has given yet another method and determination for the
( )(0).
6. The Stieltjes constants
The Stieltjes constants An = (-1)n y/n! are the Laurent coefficients for c(s)
(cf. (3)). This definition differs from Stieltjes's original usage in which the constants
yn were investigated. The definition followed here agrees with Liang & Todd
(1972), Apostol (1985) and Ivi6 (1985). There appears to be no general agreement
on the terminology.
For a proof of this assertion see Briggs & Chawla (1955) or Ivi6 (1985). Two variants
of the definition, both derived from Theorem 3 (cf. ?3), are:
n
y = lim (- 1)n(n)() n +!()
ln.imra A (
s l s -l I (1 -s )n+
(26)
co
n =-2 Im{(l + iy)-llnn(l + iy))
dy, n > 0.
Jon e27cy _I
This follows from the observation that yn is the Euler-Maclaurin constant of the
functionf(x) ==(lnn x)/x. The last formula appears suitable for deriving an asymptotic
estimate for large n. One of the earliest compilations of the constants was made by
Gram (1895) who improved Jensen's calculation and gave An to n = 15. The Stieltjes
constants also appear in Berndt ((ed.) 1985). Liang & Todd (1972) published the first
twenty yn to 15 significant figures (for 78 and y9 the last digits are incorrect). More
recent computations are reported by Bohman & Froberg (1988), Keiper (1992) and
Dilcher (1992). The apparent irregularity of the signs of the constants has been
investigated by Briggs (1955), Mitrovic (1962), Bohman & Froberg (1988). Matsuoka
(1989) has investigated the asymptotic behaviour of yn.
For the error estimate Liang & Todd argue that since
f(r)(x) will be of one sign for x > xo(r), where xo(r) = O(r). In this case xo(r) must be
estimated for each n and r. Other than this difficulty, xo(r) in practice turns out to
be large so that large numbers of terms in (29) will be summed. In the actual
Interchanging the order of summation and integration (Knopp 1964, p. 351) we have
n k lnk-j a
!
IRm(Go)I O 4(2m--1)!
(27ta)2m k= _O 1 --
(ki j)ak, 2m (2m)j , 1 < 0 <
2.
Thus the remainder is of the same order of magnitude as the last omitted term. We
are now free to choose a. We do not take a too small, for then the terms containing
the B2s in (29) diverge prematurely. In our actual computation both a, m were varied
to minimize the remainder. In general a = 10 was adequate. This is a significant
reduction in the number of terms to be summed, thus reducing rounding errors. To
give one numerical example we take n = 2, a = 10, m = 30, and let 52 be the
computed value. The error estimate shows 1R30(oo0)1 2.8 x 1027, the term neglected
is - -1.4 x 10-27, whereas the actual error is ly?2-i21 = 3.2 x 10-28. The actual
error is less than the error estimate and Y2 is correct to 25 significant figures. As
another example we take n = 17, a = 10, m = 20. The error estimate shows
_
1R20(oo)1< 2.0 x 10-23, the term neglected is 1.0 x 10-23, whereas the actual error
is ly17-7 717 = 1.5 x 10-26. Again the error estimate is very good.
The Stieltjes constants are difficult to compute with precision. To see this, we take
a = 10, n = 10, 20 and compare the four terms on the right side of (29) with yn:
n = 10: 679.2 + 209.5-877.0- 1.7->y10 2.05(-4),
n = 20: 1.1588(6)+0.8776(6)-1.9245(6)-0.1119(6) ->Y20 4.66(-4).
Hence there are severe cancellations. With increasing n this only becomes worse.
Since a > 1, logl a In a, so there is some advantage in using the transformation
ln a = M-1 log10a, and factoring out M-n. Here M = (In 10)-1 = 0.434 29 .... Indeed,
when a = 10, the logarithmic terms in all but the first summation term drop out, a
nice feature. Although this helps to gain a few extra digits, the best solution appears
to be multiple-precision arithmetic. Even without the benefit of any such package we
calculated y20 correctly to 20 significant figures, and y50 to 13 significant figures. Only
in one case, for n = 98, a ten-figure accuracy was not achieved. The last two digits
of 798, namely 50, shown underlined in table 5, are from Keiper. In our calculation
we have 32 which we suspect to be in error. The entries in table 5 should be correct
to the figures given. Compare this with Bohman & Fr6berg (1988), who using
multiple precision (33 significant figures) obtained /20 correctly to 20 significant
figures, but their y 0 = 126.95... is correct only to three significant figures. Thus
Bohman & Fr6berg's method appears to break down for large n, and the accuracy of
their table 2 is questionable. Incidentally, neither of the two authors quoted above
give any error analysis. The entries in table 5 were checked against a table of yn that
was kindly supplied by Dr J. B. Keiper.
Proc. R. Soc. Lond. A (1995)
kk k
-10 -0.41722804076736685681 -0.74477320813305334179
-9 - 0.409 044 97151 4 698 82516 - 0.728 262 005 227 376 535 25
-8 -0.39906966894504503551 -0.708305184096473476 84
-7 -0.38664273176530440937 -0.68370495235938298928
-6 -0.370737 657 20476201404 -0.652 640675963 713 78003
-5 -0.34966628059831413714 -0.61220978517348881833
-4 -0.320451 26422857728279 - 0.557504216 966569 43837
-3 -0.277 343047 840129526 98 -0.479599413658807 37989
-2 - 0.207 886 224 977 354 566 02 - 0.360 854 339 599 947 607 35
-1 -0.08333333333333333333 -0.16542114370045092921
2 -1.46035450880958681289 -3.92264613920915172747
3 -0.97336024835078271547 -2.17130135211640157070
4 -0.81327840526189165652 - 1.69844515038963403383
5 -0.733 920 924 896 340 592 24 - 1.482 800 203 258 094 635 26
6 -0.68658158194735822784 -1.360061 148668291 15548
7 -0.65515355797159740930 - 1.28100437948720628079
8 -0.632775623498695 255 29 - 1.22589140479432235432
9 -0.616033466461 17761655 - 1.18529822668961486181
10 -0.603037 519 856 241715 25 - 1.154165154109 753 987 56
The similarity between (3) and (31) suggests some possible relation between n) (0)
and y.n For n = 0 and 1, (0) = -, -'(0) = '1n(2n). Ramanujan (Berndt (ed.)
1985) gave the next relation:
(2)(0) = - iln2(2z)-_i2 2+12 + Y1.
Apostol (1985) has investigated the general relationship fully. Following his method
we give below two more explicit relations:
Table 4 (cont.)
5.3 - 2.234490 342 443 016 996 8( -2) - 3.036 197 996 697 130408 9(-3)
5.4 -2.061 389607264127021 3(- 2) -3.728608206892631 8722(-3)
5.5 - 1.902 833 981333 391 0684(-2) -4.339 595413423215 3680( -3)
5.6 - 1.757458414069001 8256(-2) - 4.860 252 061465461851 5(- 3)
5.7 - 1.6240442145295806431(-2) - 5.282 705 697 301897 809 8(- 3)
5.8 - 1.501500 995 035 807 142 2( -2) -5.600219 366 834 105 3264( -3)
5.9 - 1r3888511564138771269(-2) -5.8072922938175543558( -3)
6.0 - 1.2852165131795725076(-2) -5.899759 1435159374506( -3)
6.1 - 1.189806 727 119596 707 2(- 2) -5.874886043 095 801 8329(-3)
6.2 - 1.101909274509 794531 0(-2) - 5.731461396411 730 242 2( - 3)
6.3 - 1.020 880 718 382 939 936 9(-2) - 5.469 879 401204 083 416 4( - 3)
6.4 - 9.461390 949 627 629 202 2(- 3) -5.092 214054 786 382 258 3( -3)
6.5 -8.771 572542939033 3889(-3) -4.602281324001 5708508( -3)
6.6 -8.134570205413 351 6308( - 3) -4.005687 060871 8060865(- 3)
6.7 - 7.546 040 584 535 372 296 2(- 3) -3.309858171584237 6700(-3)
6.8 - 7.002033499128 221222 7(- 3) - 2.524054498198 533 285 8( - 3)
6.9 - 6.498951989 976 394 122 7(- 3) - 1.659358 854987 563 242 3( -3)
7.0 - 6.033 516 960 875 637 794 0( - 3) - 7.286426 801 592 406524 7( -4)
7.1 -5.602 735 816 8045654084(- 3) 2.534951753903837 1359( -4)
7.2 - 5.203 874 591091080 765 2( - 3) 1.270 818 891754 827 693 8(- 3)
7.3 -4.834433 125 2228036492(- 3) 2.305 577 562 704 955 232 5(- 3)
7.4 - 4.492 122 925 609 989 794 6(- 3) 3.338660501 0844595267( - 3)
7.5 -4.174847373018551039 1(-3) 4.349 787 661 844 826 824 3( -3)
7.6 - 3.880684004079521881 9( -3) 5.317 736517368 1182939( -3)
7.7 - 3.607 868 621 507 5522982(-3) 6.220 606 282 698 379 353 5( - 3)
7.8 -3.354781021462248 1766(-3) 7.036 119 869 888241 781 8(- 3)
7.9 -3.119932 153 720 806 7983(-3) 7.741 963 347 007 439469 8(- 3)
8.0 -2.901 952553710673 1304(-3) 8.316 161 985602247 359 5( -3)
8.1 -2.699 581 905 569 595 719 0(-3) 8.737491 1989392704345(-3)
8.2 -2.511659612752 1004929(-3) 8.985919802133401 2592( -3)
8.3 - 2.337116 267 699 880 905 ( - 3) 9.043 082 065 975 859 720 6( - 3)
8.4 -2.174965925086 1122127( -3) 8.892 773 992 785 824 937 4( -3)
8.5 - 2.024 299 094 422 239 007 2(- 3) 8.521468121 1852583502(-3)
8.6 - 1.884 276 377 626 553 953 2( - 3) 7.918839976385091 4566( -3)
8.7 - 1.754122 685 704955 303 5( -3) 7.078 298 035 520 054 289 6( - 3)
8.8 - 1.633121976 616906225 7( - 3) 5.997 507 789 366 679 734 9(- 3)
8.9 - 1.5206124592934796333(-3) 4.678899 171 777465 892 2(-3)
9.0 - 1.415 982 227 241809 1050( - 3) 3.130 1453197885727549( -3)
9.1 - 1.318665264 719 881686 1(- 3) 1.364 599 348 380 618 336 7( - 3)
9.2 - 1.228 137 804 740210 1906(- 3) -5.983253933577946521 1(-4)
9.3 - 1.1439149965799852502(-3) - 2.732 847 236 296 1190562( -3)
9.4 - 1.065547 856657 0229191(- 3) - 5.006 596 904 778 063 7890(- 3)
9.5 -9.926 204 760 338 182 1399(-4) - 7.380504448 811 925 1305( - 3)
9.6 -9.247 474609617 867 1601(-4) - 9.808810 365056465 760 8( - 3)
9.7 - 8.615 715 852 679035 829 3( -4) - 1.2239219132519115755(- 2)
9.8 -8.027616355091072221 8(-4) - 1.461321 3289522590744(--2)
9.9 -7.480 104317086533497 3( -4) - 1.686654566047 3143895(- 2)
10.0 -6.970330081 7139369368(-4) - 1.892992633814037 4229(-2)
10.1 -6.495649403 763 1530525(-4) -2.072991947182611 7538( -2)
10.2 - 6.053 608 052 860 290 429 7( - 4) -2.219006 273 600 895 671 2( -2)
10.3 -5.641927 636129 146 064 8( -4) -2.323221949123449771 7( -2)
10.4 - 5.258 492 536 566 246 910 6( - 4) -2.377 817 003 593 683 275 9( -2)
10.5 -4.901337 872 918 894082 4(-4) -2.375144392688431 9067( -2)
Proc. R. Soc. Lond. A (1995)
Table 4 (cont.)
10.6 -4.568 638 395515 430 893 8( -4) - 2.307 939 006 764 253 380 8( - 2)
10.7 - 4.258 698 240 283 856 6513( - 4) - 2.169 547 509 067 880 7554( - 2)
10.8 - 3.969941470205 265 3362( -4) - 1.954 1793465947138198( -2)
10.9 - 3.700 903 339 767 202 470 3( - 4) - 1.657 176475178674 162 8(- 2)
11.0 - 3.450 222 223 683 634 969 3( -4) - 1.275 298447 996 665611 4( -2)
11.1 -3.216632 1562987662453( -4) - 8.070185 379 761285 327 7( - 3)
11.2 -2.998 955 932 749 681 332 3( -4) -2.528 255071 584 197 263 0( -3)
11.3 -2.796098727 1793457785(-4) 3.844 754 889 170 282 110 9(- 3)
11.4 - 2.607 042187 112 503 943 0( -4) 1.099 470 550 960 544 995 8(- 2)
11.5 - 2.430 838 966 573 070 963 9( - 4) 1.883 862 138459 649 956 5(-2)
11.6 -2.266 607 663 668 706 440 1( -4). 2.726221740338291 3890( -2)
11.7 -2.1135281312284353571 (-4) 3.611 781 4223805168890( -2)
11.8 - 1.970837131680976826 1( -4) 4.522 276 693 305 724 079 7( - 2)
11.9 - 1.837 824 309 730297 3489(-4) 5.435854161 7167766875(-2)
12.0 - 1.7138284585435376683( -4) 6.327 058 334 146 300 059 5( -2)
12.1 - 1.598234057 135 173 989 8( -4) 7.166912574087 5946224( -2)
12.2 - 1.490468 058428 233 799 8( -4) 7.923 109 640 680 547 508 7( -2)
12.3 - 1.389996909114870 1043( -4) 8.560 327 284 863 866 149 3( -2)
12.4 - 1.296 323 783 939 203 550 2(- 4) 9.040 684 025 846 093 506 3( - 2)
12.5 - 1.208986018398 1762980( -4) 9.324 349 382 377 824 611 2( - 2)
12.6 - 1.1275527251130053160(-4) 9.370321 408540417 2283( -2)
12.7 - 1.051 62258027275275782 8(-4) 9.137 3822975146346088( -2)
12.8 -9.808217676273 248411 2( -5) 8.585 239 983 881 906 0203( -2)
12.9 - 9.148 020 684 049 400 626 9( - 5) 7.675860012686631793 1( -2)
13.0 - 8.532 390 865 593 045 869 9( - 5) 6.374987 374457 688028 6( - 2)
13.1 - 7.958 305 993 915 804 266 2( - 5) 4.653 852 461440 004 332 7( -2)
13.2 - 7.422 950 244 838 794 4219( - 5) 2.491048726522 1526192( -2)
13.3 - 6.923 699 945 003 723 472 8( - 5) - 1.2543801396709190533(-3)
13.4 - 6.458110 320671 339097 2(- 5) -3.196077412343427 8073( -2)
13.5 - 6.023 903175 237 361566 6(- 5) -6.707 558 118086 077 548 4( -2)
13.6 -5.618955428778635571 1(-5) - 1.063051635678831 9420( - 1)
13.7 - 5.241288 457 902 249 975 4( - 5) - 1.491 7290618631251487( - 1)
13.8 -4.889058178743 1345867( -5) - 1.949 977 598 480 229 001 3( -1)
13.9 -4.560545820173334 1963( -5) -2.4287464646038180933(- 1)
14.0 -4.254 149 338 178 082 305 4( - 5) - 2.916577 247 438 735 203 2(- 1)
14.1 -3.968375425946651 8620( -5) -3.399 476 474 667 578 968 9( - 1)
14.2 -3.7018320775441001196(-5) - 3.8608406852410199687( - 1)
14.3 - 3.453 221666095 675 073 6( -5) - 4.281447 752 668 146 832 8(- 1)
14.4 - 3.221334 500 2491 11641 5( -5) -4.639529221 1577068597(- 1)
14.5 - 3.005 042 825 299 870 732 9( - 5) -4.910939185023 732 804 1(- 1)
14.6 -2.803 295 237 787533510 1( -5) - 5.069 435 693 811 357 670 5( - 1)
14.7 -2.615111 484613588078 1(-5) -5.087 090 707 938 058 1684( - 1)
14.8 -2.439577 619805 961 0423( -5) -4.934 844 161229 039 822 7( -- 1)
14.9 - 2.275 841493 976 887 539 5( -5) -4.583 216598 108 368 365 8( - 1)
15.0 - 2.123108 553 300034 139 3( - 5) -4.003193028077 255 9384(- 1)
15.1 -1.980637926481 1574652(-5) -3.1672879574502493889(- 1)
15.2 - 1.847 738 779 724067 057 3(-5) - 2.050 797 895 307 672 134 7(- 1)
15.3 - 1.723766921 1095138546(- 5) -6.332428663 721 7432699( -2)
15.4 - 1.608 121637 1173474330( -5) 1.1000075162482746758(- 1)
15.5 - 1.500242745239694023 2(-5) 3.155 935 162 754465 7754(- 1)
15.6 - 1.399 607 847 762196 729 8(-5) 5.531 922 244 659 889 820 7( - 1)
15.7 - 1.305 729 772 838151096 8( -5) 8.213678 1546084295796( - 1)
15.8 - 1.218 154189952 761 855 7(-5) 1.117 308 691264 306039 7(0)
Proc. R. Soc. Lond. A (:1995)
Table 4 (cont.)
8
7. The tables
For integer arguments the zeta-function has been tabulated extensively (Stieltjes
1887; Glaisher 1914; Lienard 1948). See Fletcher et al. (1962) and Davis (1963) for
the review of the various tables. Stieltjes gives Y(n) to 32 decimal places for
n = 2(1)70. Glaisher extended this to n = 107; the table is reprinted in Davis (1963,
II). Among recent works McLellan (1968) and Morris (1973) deserve special mention.
n Yn n Yn
21 0.104 437 769 756 (-3) 51 -0.191 969 118 7 (+2)
22 -0.541 599 582 204 (-3) 52 -0.463 188 923 0 (+3)
23 -0.124 396 209 041 (-2) 53 -0.134 065 914 4 (+4)
24 -0.158 851 127 890 (-2) 54 -0.257 245 474 0 (+4)
25 -0.107 459 195 274 (-2) 55 -0.345 714 120 9 (+4)
26 0.656 803 518 637 (-3) 56 -0.205 527 581 6 (+4)
27 0.347 783 691 362 (-2) 57 0.537 228 221 3 (+4)
28 0.640 006 853 170 (-2) 58 0.240 193 893 8 (+ 5)
29 0.737 115 177 047 (-2) 59 0.574 243 193 0 (+ 5)
30 0.355 772 885 557 (-2) 60 0.985 432 545 9 (+ 5)
31 -0.751 332 599 782 (-2) 61 0.111 670957 8 (+6)
32 -0.257 037 291 084 (-1) 62 0.533 366 521 1 (+4)
33 -0.451 067 341 081 (-1) 63 -0.390 972 687 3 (+6)
34 -0.511 269 280 215 (-1) 64 -0.130 318 071 3 (+7)
35 -0.203 730 436 039 (-1) 65 -0.284 507 655 3 (+ 7)
36 0.724 821 588 168 (-1) 66 -0.454 052 661 0 (+ 7)
37 0.236 026 382 274 (0) 67 -0.434 190 513 9 (+ 7)
38 0.428 963 446 385 (0) 68 0.287 156 694 6 (+ 7)
39 0.517 921 842 693 (0) 69 0.266 049 085 5 (+ 8)
40 0.248 721 559 395 (0) 70 0.793 216 631 2 (+8)
41 -0.719 574 846 901 (0) 71 0.166 215 1340 (+9)
42 -0.263 879 492 734 (+1) 72 0.255 153 258 3 (+9)
43 -0.526 493 031 236 (+1) 73 0.212 655 631 7 (+9)
44 -0.718 874 588 950 (+1) 74 -0.298 767 089 4 (+ 9)
45 -0.507 234 458 992 (+1) 75 -0.191 948 742 8 (+ 10)
46 0.660 991 560 910 (+1) 76 -0.551 557 425 8 (+10)
47 0.340 397 749 822 (+2) 77 -0.114 834 509 9 (+11)
48 0.786 824 797 632 (+2) 78 -0.175 701 522 8 (+ 11)
49 0.125 844 387 632 (+3) 79 -0.139 610 214 6 (+11)
50 0.126 823 602 651 (+3) 80 0.251 634410 1 (+ 11)
Table 5. (cont.)
n Yn n Yn
81 0.151058 510 8 (+12) 91 -0.147 797 700 1 (+15)
82 0.437 904 431 2 (+12) 92 0.129 463 214 1 (+15)
83 0.931 706 846 9 (+12) 93 0.118 856 297 6 (+16)
84 0.147 209 981 9 (+13) 94 0.392 068 462 7 (+16)
85 0.125 904 496 8 (+13) 95 0.934 167 0850 (+16)
86 -0.195 881022 5 (+13) 96 0.170 752 474 5 (+17)
87 -0.129 515 455 0 (+13) 97 0.207 179 835 4 (+17)
88 -0.392 971 538 8 (+14) 98 -0.285 430 785 0 (+16)
89 -0.875 304 474 0 (+14) 99 -0.112 584 810 8 (+18)
90 -0.147 161 049 4 (+15) 100 -0.425 340 157 2 (+18)
See also Moshier (1989), Piessens & Branders (1972), Cody et al. (1971) and Fransen
& Wrigge (1980).
It is surprising that very few tables of '(s) have been published. Occasionally short
tables of '(s)/~(s) have appeared in papers devoted to number theory. One of the
earliest such tables is by Walther (1926). Walther gave '(s)/S(s) for s = -6(0.1)7 to
seven decimal places. For a long time Walther's table has been widely reprinted and
cited (see, for example, Dwight 1958; also Lebedev & Fedorova 1960; Fletcher et al.
1962). However, we noticed 15 instances of disagreement, mostly in the last digits;
twelve of these are for the negative arguments. Rosser & Schoenfeld (1962) tabulated
'(n), '(n)/~(n) for n = 2(1)29 to 17 decimal places. Our calculation generally agrees
with Rosser except for the entry ~'(6), where in place of Rosser's 2 as the last digit,
we have 3. For '(n)/,(n) there were six additional disagreements for n = 2, 8, 11, 21,
22 and 23. Only a selection of our computation is given here. Table 3 gives ~(l/k),
'(1l/k) for k =- 10(1)10. A short table of '(s) for s = -20(0.1)20 appears in table 4.
Because the Stieltjes constants appear in many of the formulas, and these
fundamental constants are extremely difficult to evaluate, we give in table 5 the first
one hundred Yn.
References
Apostol, T. M. 1985 Formulas for higher derivatives of the Riemann zeta function. Math. Comp.
44, 223-232.
Ayoub, R. 1974 Euler and the zeta function. Am. math. Mthly 81, 1067-1086.
Bendersky, L. 1933 Sur la fonction gamma generalisee. Acta math. 61, 263-322.
Berndt, B. C. 1972 On the Hurwitz zeta-function. Rocky Mtn J. Math. 2, 151-157.
Berndt, B. C. (ed.) 1985 Ramanujan's notebooks,part I. New York: Springer-Verlag.
Bohman, J. & Fr6berg, C.-E. 1988 The Stieltjes function - definition and properties. Math. Comp.
51,281-289.
Briggs, W. E. 1955 Some constants associated with the Riemann zeta-function. Michigan math. J.
3, 117-121.
Briggs, W. E. & Chawla, S. 1955 The power series coefficients of (s). Am. math. Mthly 62, 323-325.
Cody, W. J., Hillstrom, K. E. & Thacher, H. C. Jr 1971 Chebyshev approximations for the
Riemann zeta function. Math. Comp. 25, 537-547.
Davis, H. T. 1962 The summation of series. San Antonio, Texas: The Principia Press of Trinity
University.
Davis, H. T. 1963 Tables of the mathematicalfunctions, vols I and ii (revised and enlarged edn). San
Antonio, Texas: The Principia Press of Trinity University.
Proc.R. Soc.Lond.A (1995)