Professional Documents
Culture Documents
1001 2835
1001 2835
Donal F. Connon
16 January 2010
Abstract
In a rather straightforward manner, we develop the well-known formula for the Stirling
numbers of the first kind in terms of the (exponential) complete Bell polynomials where
the arguments include the generalised harmonic numbers.
We also show how the (exponential) complete Bell polynomials feature in a number of
other areas of mathematical interest.
EXPLICIT FORMULA FOR THE STIRLING NUMBERS OF THE FIRST KIND
The Stirling numbers s (n, k ) of the first kind [23, p.56] are defined by the following
n
generating function (the bracket symbol is also employed)
k
(1.1)
x( x 1) ( x n + 1) = s (n, k ) x k
k =0
x( x 1) ( x n + 1) = s (n, k ) x k
k =0
k =0
k =1
We also consider the ascending factorial symbol ( x ) n , also known as the Pochhamer
symbol, defined by [23, p.16] as
(1.4)
( x) n = x( x + 1)( x + 2) ( x + n 1) if n > 0
( x)0 = 1
x( x + 1)( x + 2) ( x + n 1) ( x) = (n + x)
Hence we have
(1.5)
( x) n =
(n + x) x ( n + x)
=
( x)
(1 + x)
For the reasons discussed below, we shall find it more convenient to deal with the
function ( x) n divided by x . We consider the first derivative
(1.6)
d ( x) n (n + x)
=
[ (n + x) (1 + x)]
dx x
(1 + x)
( x) =
d
log ( x)
dx
By definition we have
(n + x) (1 + x) =
d
d
log (n + x) log (1 + x)
dx
dx
d
( n + x)
log
dx
(1 + x)
( x)
d
log n
dx
x
d
1
log( x) n
dx
x
( x)n 1
1
1
= +
+ +
= H n(1) ( x)
( x)n x x + 1
x + n 1
where H n( m ) ( x) is the generalised harmonic number function defined by
(1.7)
n 1
1
m
k =0 (k + x)
H n( m ) ( x) =
(m)
n
(1) = H
(m)
n
1
m
k =1 k
n
(n + x) (1 + x) = H n(1) ( x)
1
x
n 1
1
k =1 k + x
=
n 1
1
.
k =1 k + x
and we denote g ( x) =
This gives us the relationship (which in turn leads us to consider the (exponential)
complete Bell polynomials)
d ( x)n ( x)n
=
[ (n + x) (1 + x)]
dx x
x
(1.8)
( x) n
g ( x)
x
We see that
g (0) = H n(1)1
n 1
1
r +1
k =1 ( k + x )
g ( r ) ( x) = (1) r r !
(1.9)
d r f ( x)
e
= e f ( x )Yr ( f (1) ( x), f (2) ( x),..., f ( r ) ( x) )
r
dx
where the (exponential) complete Bell polynomials may be defined by Y0 = 1 and for
r 1
k1
k2
kr
r!
x1 x2
xr
Yr ( x1 ,..., xr ) =
(1.10)
...
r!
( r ) k1 ! k 2 !... k r ! 1! 2!
where the sum is taken over all partitions (r ) of r , i.e. over all sets of integers k j such
that
The complete Bell polynomials have integer coefficients and the first six are set out
below (Comtet [6, p.307])
(1.11)
Y1 ( x1 ) = x1
Y2 ( x1 , x2 ) = x12 + x2
Y3 ( x1 , x2 , x3 ) = x13 + 3x1 x2 + x3
Y4 ( x1 , x2 , x3 , x4 ) = x14 + 6 x12 x2 + 4 x1 x3 + 3x22 + x4
Y5 ( x1 , x2 , x3 , x4 , x5 ) = x15 + 10 x13 x2 + 10 x12 x3 + 15 x1 x22 + 5 x1 x4 + 10 x2 x3 + x5
Y6 ( x1 , x2 , x3 , x4 , x5 , x6 ) = x16 + 6 x1 x5 + 15 x2 x4 + 10 x23 + 15 x12 x4 + 15 x23 + 60 x1 x2 x3
tj
tn
exp x j = Yn ( x1 ,..., xn )
n!
j =1 j ! n =0
(1.12)
and hence we note that (1.12) is simply the corresponding Maclaurin series.
We note that
tn
tj
tj
t j
Y
ax
ax
ax
a
x
x
(
,...,
)
exp
exp
exp
=
=
=
j j!
j
j
1
n
n
n!
n =0
j =1
j =1 j !
j =1 j !
tn
tn
Y
x
x
Y
ax
ax
(
,...,
)
(
,...,
)
=
n
n
n 1
n 1
n !
n!
n =0
n =0
Let us now consider a function f (t ) which has a Taylor series expansion around x : we
have
tj
tj
e f ( x +t ) = exp f ( j ) ( x) = e f ( x ) exp f ( j ) ( x)
j!
j!
j =0
j =1
tn
= e f ( x ) 1 + Yn ( f (1) ( x), f (2) ( x),..., f ( n ) ( x) )
n !
n =1
We see that
d m f ( x ) m f ( x +t )
m f ( x +t )
e
= me
= me
dx m
x
t
t =0
t =0
d r f ( x)
= e f ( x )Yr ( f (1) ( x), f (2) ( x),..., f ( r ) ( x) )
e
r
dx
Suppose that h( x) = h( x) g ( x) and let f ( x) = log h( x) . We see that
f ( x) =
h( x)
= g ( x)
h( x )
dr
d r log h ( x )
(
)
=
= h( x)Yr ( g ( x), g (1) ( x),..., g ( r 1) ( x) )
h
x
e
r
r
dx
dx
In particular we have
dr
h( x) = h(0)Yr ( g (0), g (1) (0),..., g ( r 1) (0) )
r
dx
x =0
h( x ) =
( x)n
x
n 1
1
k =1 k + x
g ( x) =
and it is easily seen that
(1.15)
d r ( x) n
dx r x
r 1
( r 1)! H n( r1) )
= (n 1)!Yr ( H n(1)1 , 1! H n(2)
1 ,..., ( 1)
x =0
d r ( x)n
=
(1) n + k (k 1) (k r ) s (n, k ) x k 1 r
r
dx x
k =0
d r ( x)n
dx r x
= (1) n + r +1 r ! s (n, r + 1)
x=0
Therefore, equating (1.15) and (1.17), we obtain the known relationship for Stirling
numbers of the first kind for r 0
(1.18)
s (n, r + 1) = (1) n + r +1
(n 1)!
r 1
(r 1)! H n( r)1 )
Yr ( H n(1)1 , 1! H n(2)
1 ,..., ( 1)
r!
The above relationship was previously derived by Klbig [14] but this appears to be a
more direct proof of this important formula.
The first few Stirling numbers s (n, k ) of the first kind are easily determined from (1.18);
these are also reported in [21] and in the book by Srivastava and Choi [23, p.57]
(1.19)
s (n, 0) = n ,0
s (n,1) = (1) n +1 (n 1)!
s (n, 2) = (1) n (n 1)! H n(1)1
s (n,3) = (1) n +1
s (n, 4) = (1) n
{(
(n 1)!
H n(1)1
2
{(
H n(2)
1
3
(n 1)!
(3)
H n(1)1 3H n(1)1 H n(2)
1 + 2 H n1
6
(r k )!
r k 1
(r k 1)! H n( r1 k ) )
s(n, r k + 1) = Yr k ( H n(1)1 , 1! H n(2)
1 ,..., ( 1)
(n 1)!
r
r
Yr +1 ( x1 ,..., xr +1 ) = Yr k ( x1 ,..., xr k ) xk +1
k =0 k
dr
d r ( n + x)
x
(
)
=
= (1) n + k k (k 1) (k r + 1) s (n, k ) x k r
n
r
r
dx
dx ( x)
k =0
(1.22)
dr
( x) n
dx r
= (1) n + r r ! s (n, r )
x =0
d
d (n + x) ( x)(n + x) ( x)(n + x)
( x) n =
=
dx
dx ( x)
2 ( x)
( n + x)
[ (n + x) ( x)]
( x )
Therefore we have
(1.23)
( x) n
d
( x) n = ( x) n H n(1) ( x) =
x H n(1) ( x)
dx
x
Hence we obtain
d
( x)n
dx
= (n 1)!
x =0
dr
( x)n
dx r
d2
mathematics is somewhat more cumbersome. For example, let us consider 2 ( x) n
dx
2
d2
( x) n = ( x) n H n(1) ( x) H n(2) ( x)
2
dx
We have by definition
.
x =0
H n(1) ( x + 1) =
1
1
1
+
+ +
x +1 x + 2
x+n
and therefore
H n(1) ( x) =
1
1
1
1
1
+
+ +
=
+ H n(1) ( x + 1)
x x +1
x + n 1 x x + n
n
+ H n(1) ( x + 1)
x ( x + n)
This gives us
2
n2
2n
(1)
H ( x) = 2
+
H
(
x
+
1)
+
H n(1) ( x + 1)
n
2
x ( x + n)
x( x + n)
2
(1)
n
Similarly we have
H n(2) ( x) =
1
1
+ H n(2) ( x + 1)
2
2
x ( x + n)
2nx + n 2
+ H n(2) ( x + 1)
2
2
x ( x + n)
This gives us
x =0
2
( x)n
x H n(1) ( x) H n(2) ( x)
x 0
x
= lim
( x)n n
1
=
z
x
(1 z )
n =0 n !
(2.1)
=
n!
(1 z ) x
n =0
(2.2)
log r (1 z ) d r
zn
x
=
(
)
n
r
(1 z ) x
n!
n = 0 dx
dr
( x)n
r
n = 0 dx
(1) r log r (1 z ) =
x =0
zn
n!
10
dr
( x)n
dx r
= (1) n + r r ! s ( n, r )
x =0
we then easily obtain the well-known Maclaurin expansion due to Cauchy [23, p.56]
(2.3)
zn
n!
zn
n!
A different proof of (2.3) was given by Pyla and Szeg in [18, p.227].
Letting x = 1 in (2.2) we obtain the well-known generating function for the harmonic
numbers
(2.3.1)
log(1 z ) (1) n
= Hn z
1 z
n =0
(2.4)
( x) n H n(1) ( x) H n(2) ( x)
n!
)z
2
log 2 (1 z )
= H n(1) H n(2) z n
1 z
n =0
Such series are considered in more detail in, for example, [7] and the references
contained therein.
Integrating (2.5) results in
1 3
z n +1
(1) 2
(2 )
log (1 z ) = H n H n
3
n +1
n =0
2
(1)
1
1 z n +1
(2)
= H n +1
H n +1 +
n + 1
(n + 1) 2 n + 1
n =0
11
2 H n(1)+1
2 z n +1
= H n(1)+1
H n(2)
+
+1
n +1
(n + 1) 2 n + 1
n =0
2 H n(1)
2 zn
= H n(1)
H n(2) + 2
n
n n
n =1
We therefore have
(2.6)
H n(1) n
zn
1 3
(1) 2
(2)
log (1 z ) = H n H n
2 2 z + 2 Li3 ( z )
3
n
n =1
n =1 n
zn
Lis ( z ) = s
n =1 n
(2.7)
zn
n!
2
zn
= 3 H n(1)1 H n(2)
1
n
n =1
(1) 2 2 H n(1)
2 zn
(2)
= 3 H n
Hn + 2
n
n n
n =1
f ( x) =
n
n (1) k
n!
=
x( x + 1)...( x + n) k =0 k k + x
n + x
=x
n
1
, x {0, 1,..., n}
12
n!
(n + 1)
=
x(1 + x)...(n + x)
( x) n +1
Differentiation results in
f ( x) =
(n + 1) (1)
H n +1 ( x)
( x) n +1
(n + 1)
r
(r )
Yr ( H n(1)+1 ( x),1! H n(2)
+1 ( x ),..., ( 1) ( r 1)! H n +1 ( x ) )
( x) n +1
n (1) k
(n + 1) (1) r
r
(r )
=
Yr ( 0! H n(1)+1 ( x),1! H n(2)
+1 ( x ),..., ( 1) ( r 1)! H n +1 ( x ) )
r +1
( x) n +1 r !
k = 0 k (k + x)
n
It may be noted that Coppo [10] has expressed this in a slightly different form
(3.3)
n (1) k
(n + 1) 1
(r )
=
Yr ( 0! H n(1)+1 ( x),1! H n(2)
+1 ( x ),..., ( r 1)! H n +1 ( x ) )
r +1
( x) n +1 r !
k =0 k (k + x)
n
and reference to (1.10.2) shows that these are equivalent statements. It may be noted that
n
n (1) k
reference to the right-hand side of (3.3) shows that
is positive for x > 0
r +1
k =0 k (k + x)
(which otherwise does not appear to be immediately obvious).
Particular cases of Coppos formula are set out below.
13
(3.4)
n (1) k
k ( k + x) =
k =0
(n + 1)( x)
(n + 1)( x)
Y0 =
(n + 1 + x)
( n + 1 + x)
n (1) k
(n + 1)( x)
( n + 1)( x) (1)
=
Y1 H n(1)+1 ( x) =
H n +1 ( x)
2
( n + 1 + x)
( n + 1 + x)
k =0 k (k + x)
(3.5)
n (1) k
(n + 1)( x) 1
=
Y2 H n(1)+1 ( x), H n(2)
+1 ( x )
3
(n + 1 + x) 2
k =0 k (k + x)
2
(n + 1)( x) 1
H n(1)+1 ( x) + H n(2)
+1 ( x )
( n + 1 + x) 2
(3.6)
n
n (1) k
k ( k + x)
k =0
(3.7)
(n + 1)( x) 1
(3)
Y3 H n(1)+1 ( x), H n(2)
+1 ( x ), 2 H n +1 ( x )
(n + 1 + x) 6
3
(n + 1)( x) 1
(3)
H n(1)+1 ( x) + 3H n(1)+1 ( x) H n(2)
+1 ( x ) + 2 H n +1 ( x )
( n + 1 + x) 6
n (1) k
We now wish to consider representations for r . We see that
k =1 k k
n
n
n (1) k
n (1) k 1 (n + 1)( x) 1
=
=
x (n + 1 + x) x
k =1 k k + x
k =0 k k + x
n
(n + 1) x( x) (n + 1 + x)
x(n + 1 + x)
(n + 1)(1 + x) (n + 1 + x)
x(n + 1 + x)
x 0
0
k
x(n + 1 + x)
k =1 k
n
14
(n + 1)(1 + x) (n + 1 + x)
x0
x(n + 1 + x) + (n + 1 + x)
= lim
(n + 1)(1) (n + 1)
(n + 1)
= (1) (n + 1)
= H n(1)
Hence we obtain Eulers well-known identity
n (1) k
= H n(1)
k
k =1 k
n
(3.8)
2
x
(n + 1 + x)
k =1 k ( k + x )
n
(n + 1) x 2 ( x) H n(1)+1 ( x) (n + 1 + x)
x 2 (n + 1 + x)
(n + 1) (1 + x) x H n(1)+1 ( x) (n + 1 + x)
=
x 2 (n + 1 + x)
and since lim x H n(1) ( x) = 1 we see that we may apply LHpitals rule to obtain
x 0
n (1) k
2
k =1 k k
n
(1)
(1)
Since lim[ H n(1)+1 ( x) x H n(2)
we have
+1 ( x )] = H n
x 0
(n + 1) H n(1) + (n + 1)(1) (n + 1)
0
15
(n + 1) H n(1) + (1) (n + 1)
0
0
0
= lim
x 0
lim
x 0
(n + 1 + x)
x (n + 1 + x) + 2 x(n + 1 + x) + 2 x(n + 1 + x) + 2(n + 1 + x)
2
1
1
(n + 1)
= [2 H n(2) + 2(1) H n(1) ] (1)
=L
2
2
(n + 1)
Differentiation of (1.6.1) gives us
( x) =
so that
( x)( x) ( ( x) )
2 ( x)
= (2, x)
( x)
= 2 ( x) + (2, x)
( x)
(1)
(n + 1)
= (2) (2, n + 1) + 2 (1) 2 (n + 1)
(n + 1)
From the definition of the Hurwitz zeta function it readily follows that
( s, n + x) = ( s, x) H n( s ) ( x)
and in particular we have
Therefore we obtain
(1)
and thus
L=
2
(n + 1)
= H n(2) H n(1) + 2 2 (1) 2 (1) (n + 1)
(n + 1)
2
1
H n(2) H n(1) + 2 2 (1) 2 (1) (n + 1) + 2 (1) H n(1)
2
which simplifies to
L=
2
1
H n(2) + H n(1)
2
so that we obtain
2
n (1) k
1
= H n(2) + H n(1)
2
2
k =1 k
n
(3.9)
The above result is well known; see for example the paper by Flajolet and Sedgewick
[12] where they also reported the following identity
n (1) k 1 (1)
S n (3) = 3 = H n
6
k =1 k k
n
1
1
+ H n(1) H n(2) + H n(3)
2
3
which may also be derived by the above limiting process (albeit with much more tedious
labour).
17
Defining S n (r ) by
n (1) k
S n (r ) = r
k =1 k k
n
(3.10)
then Flajolet and Sedgewick [12] showed that S n (r ) can be expressed in terms of the
generalised harmonic numbers as
(3.11) Sn (r ) =
m1
1m1 + 2 m2 + 3 m3 ...= r
m2
m3
m1 ! m2 ! m3 !...mr ! 1 2 3
r
mr
1
Yr 0! H n(1) ,1! H n(2) ,..., (r 1)! H n( r )
r!
(3.12)
+1 (1 + x ),..., ( 1) ( r 1)! H n +1 (1 + x ) )
r +1
(1 + x) n +1 r !
k =0 k (k + 1 + x)
n
+1 ,..., ( 1) ( r 1)! H n +1 )
r +1
n +1 r !
k = 0 k ( k + 1)
n
r +1
n r!
k = 0 k ( k + 1)
n 1
Reindexing gives us
18
r +1
n r!
m =1 m 1 m
n
Since
(3.13)
1 n 1 1 n
= this becomes
m m 1 n m
n (1) m +1 (1) r
Yr ( 0! H n(1) ,1! H n(2) ,..., (1) r (r 1)! H n( r ) )
=
r
!
m
r
m =1
n (1) m +1 1
= Yr ( 0! H n(1) ,1! H n(2) ,..., (r 1)! H n( r ) )
r
m
r!
m =1 m
n
r
n (1) k +1
(1)
(2)
(r ) x
= Yr ( 0! H n ,1! H n ,..., (r 1)! H n )
x
kr
r!
r =0
k =1 k
r =0
n
H n( r ) r
n (1) k +1
r
=
exp
x
x
kr
r =0
k =1 k
r =1 r
Let
f ( x) = log (n + x) log (1 + x)
Then we have
f ( m ) (0) = ( m 1) (n) ( m 1) (1) = (1) m +1 (m 1)! H n( m1)
and we therefore have the Maclaurin expansion
(3.17)
(1) m +1 ( m ) m
H n 1 x
m
m =1
19
(1) m +1 ( m ) m
(n + x)
= exp
H n 1 x
(1 + x)(n)
m =1 m
Hence we obtain
(3.19)
(1 x)(n + 1) r n n (1) k +1
= x
(n + 1 x)
kr
r =0
k =1 k
20
m =0
(m)
n
) mx ! = exp Hm
m =1
xm
(1 x)(n + 1) r n n (1) k +1
= x
(n + 1 x)
kr
r =0
k =1 k
tr
r =0
(1) r ( x) n +1 n n (1) k
tr
(1)
(2)
(r )
r
0!
(
),1!
(
),...,
(
1)
(
1)!
(
)
=
Y
H
x
H
x
r
H
x
(
)
r
n +1
n +1
n +1
(n + 1) k =0 k (k + x) r +1 r =0
r!
(r )
r H n +1 ( x ) r
= exp (1)
t
r
r =1
r
k
t 1 1 t
n =1
k =1 k
(3.21)
(3.22)
xk
r
k =1 k
Lir ( x) =
We note that
(1) k
Lir (1) = r = s (r )
k =1 k
s (r ) = (1 21 r ) (r )
Therefore with t = 1/ 2 in (3.22) we obtain
21
1 1
2 Lir ( 1) = n Yr 0! H n(1) ,1! H n(2) ,..., (r 1)! H n( r )
r ! n =1 2
(r ) =
1
1
Y 0! H n(1) ,1! H n(2) ,..., (r 1)! H n( r )
n +1 r
1 r
(1 2 )r ! n =1 2
B ( x, y ) = t x 1 (1 t ) y 1 dt
0
( x ) ( y )
( x + y )
( x ) ( y )
[ ( x) ( x + y )] = t x 1 (1 t ) y 1 log t dt
B ( x, y ) =
x
( x + y )
0
(4.1)
( x ) ( n)
[ ( x) ( x + n)]
( x + n)
x 1
(1 t ) n 1 log t dt =
x 1
(1 t ) n 1 log t dt =
or equivalently
1
(4.2)
t
0
( x)(n) (1)
H n ( x)
( x + n)
With x = 1 we obtain the well known integral representation for the harmonic numbers
1
(4.3)
n (1 t ) n 1 log t dt = H n(1)
0
22
(4.4)
x 1
(1 t ) n 1 log r t dt =
( x ) ( n)
Yr 0! H n(1) ( x),1! H n(2) ( x),..., (1) r (r 1)! H n( r ) ( x )
( x + n)
(1 t )
(4.5)
n 1
1
log r t dt = Yr 0! H n(1) ,1! H n(2) ,..., (1) r (r 1)! H n( r )
n
(4.6)
The following formula was also reported by Devoto and Duke [11, p.30]
1
(1 t )
n 1
2 (2) n 1 H k(1)
log t dt = H n +
n
k =1 k + 1
2
2
k =1 k
(4.7)
1
+ H n(2)
2
As a matter of interest, I also found formula (4.7) reported by Levenson in a 1938 volume
of The American Mathematical Monthly [15] in a problem concerning the evaluation of
(4.8)
3
1
1
1
(4.9)
The following formula was also reported by Devoto and Duke [11, p.30]
1
n (1 t )
0
n 1
(1)
n 1
(3) n 1 H k(1)
H k(2) n 1 1 k 2 H j
log t dt = 6 H n +
+
+
2
k =1 ( k + 1)
k =1 k + 1
k =1 k + 1 j =1 j + 1
23
and its equivalence to (4.9) is readily seen by reference to Adamchiks formula [1]
H k( 2 ) n H k(1)
+ 2 = H n(3) + H n(1) H n( 2 )
k
k =1 k
k =1
n
(4.10)
(1) k +1 x( x 1)...( x k + 1)
k a (a + 1)...(a + k 1)
k =1
( x + a ) ( a ) =
which converges for Re ( x + a ) > 0 . According to Raina and Ladda [19], this summation
formula is due to Nrlund (see [16], [17] and also Rubens note [20]).
Then using
( x) n = x( x + 1) ( x + n 1) =
( x + n)
( x)
1 ( x)n
n =1 n ( a ) n
( a x ) ( a ) =
1 ( x) n
Yr ( H n(1) ( x), 1! H n(2) ( x),..., (1) r 1 (r 1)! H n( r ) ( x) )
n =1 n ( a ) n
(1) r ( r ) (a x) =
24
we obtain
1 ( x)n
Yr ( H n(1) (a ),1! H n(2) (a),..., (1) r ( r 1)! H n( r ) ( a ) )
n =1 n ( a ) n
(5.3) ( r ) (a x) ( r ) (a) =
As shown in [7] we may obtain numerous Euler sums by utilising these formulae.
From (4.4) we have
1
1
1
t a 1 (1 t ) n 1 log r t dt =
Yr 0! H n(1) (a),1! H n(2) (a ),..., (1) r (r 1)! H n( r ) (a )
( n) 0
(a) n
( a x )
(r )
(r )
( x) n 1 a 1
(a ) =
t (1 t ) n 1 log r t dt
n =1 n ! 0
( x) n H n(1) ( x) a 1
t (1 t ) n 1 log r t dt
n!
n =1
0
( r +1) (a x) =
and with x = 1 we have
( r +1)
(a 1) = H
n =1
(1)
n
a 1
(1 t ) n 1 log r t dt
t a 1 log r t (1)
=
H n (1 t ) n dt
1 t n =1
0
1
H
n =1
(1)
n
(1 t ) n =
log t
t
t a 2 log r +1 t
dt
1 t
0
( r +1) (a 1) =
We may express this as
25
t a 1 log r t
dt
1 t
0
(a) =
(r )
( r ) (a) = (1) r +1 r ! (r + 1, a)
which results in the well-known integral
t a 1 log r t
dt
1 t
0
(1) r r ! (r + 1, a ) =
(6.1)
bn n
x
n =1 n
log h( x) = b0 +
(6.2)
h( x) = ar x r
r =0
h( x) = h( x) bn x n 1 h( x) g ( x)
n =1
26
ar =
ar =
1 dr
h( x )
r ! dx r
x =0
1
h(0)Yr g (0), g (1) (0),..., g ( r 1) (0)
r!
We have
g ( j ) ( x) = bn (n 1)(n 2) (n j ) x n 1 j
n =1
and thus
g ( j ) (0) = j !b j +1
Therefore we obtain
ar =
=
1 b0
e Yr g (0), g (1) (0),..., g ( r 1) (0)
r!
1 b0
e Yr ( b1 ,1!b2 ,..., (r 1)!br )
r!
(6.3)
we see that
bj
h( x) = eb0 exp t j
j =1 j
bn n
x
n =1 n
log h( x) = b0 +
27
xn
n!
(6.4)
xn
n!
1
xn
= e b0 Yn ( b1 , 1!b2 ,..., (n 1)!bn )
h( x )
n!
n =0
(r + 1)ar +1 = ar k bk +1
k =0
or equivalently
r
(6.6)
rar = ar mbm
m =1
h( x)
= g ( x)
h( x )
28
xr
r!
tj
h( x) = h(0) exp g ( j 1) (0)
j!
j =1
( j 1)
j =1
tj
(0)
j!
f ( x) = f (0) + f ( j ) (0)
j =1
tj
j!
Gamma function
( n)
n=2
log (1 + x) = x + (1) n
xn
, 1 < x 1
(6.7)
) xn!
and
(6.8)
1
xn
= Yn ( 0! (1), 1! (2),..., (1) n +1 (n 1)! (n) )
n!
(1 + x) n =0
) n(n 1)...(nn! m + 1) x
29
nm
( p ) ( x) = (1) p +1 p ! ( p + 1, x)
we may express ( m ) ( x) in terms of ( x) and the Hurwitz zeta functions.
It is well known that ( x) is negative in the interval [0, ) where > 0 is the unique
solution of ( ) = 0 and hence we see that ( x) is positive in that interval. Since
Ym ( x1 ,..., xm ) > 0 when all of the arguments are positive, we deduce that ( m ) ( x) has the
same sign as (1) m when x [0, ) .
We also note that
Ym ( x1 , x2 ,..., (1) m +1 xm ) = (1)m Ym ( x1 ,..., xm )
but no discernable sign pattern emerges here. We may note that
30
dn 1
dx n h( x)
x =0
xn
(a + x) = (a ) Yn ( 0! (a ),1! (2, a ),..., (1) (n 1)! (n, a ) )
n!
n =0
n
The following identity was originally derived by Srivastava [23, p.210] in 1988
log G (1 + x) =
1
1
(k 1) k
2
log(2
1
x
(1
+
)
x
+
(1) k +1
x
[
]
2
2
k
k =3
and from this we may determines a series expansion for 1/ G ( x + 1) in terms involving the
exponential Bell polynomials. Using () we have
1
xn
= Yn ( c1 ,1!c2 ,..., (1) n (n 1)!cn )
G (1 + x) n =0
n!
where
c1 =
1
[log(2 ) 1]
2
c2 = (1 + )
cn = (n 1) for n 3
31
dm
1
= Ym ( c1 ,1!c2 ,..., (1) m (m 1)!cm )
m
dx G (1 + x) x =0
and noting (1.10.2) we determine that
dm
1
= (1) m Ym ( c1 ,1!c2 ,..., (m 1)!cm )
m
dx G (1 + x) x =0
dm
1
has the same
m
dx G (1 + x)
d
1
G(1 + x)
= 2
= G(1)
dx G (1 + x) x =0
G (1 + x) x =0
and hence we have
G(1) =
1
[log(2 ) 1]
2
Bernoulli numbers
Bn n
t
=
t
t
e 1 n =0 n !
and thus
Bn n 1
1
t
=
t
e 1 n =0 n !
Integration gives us
dt
e t dt
x
a
=
a et 1 a 1 et = log(1 e ) log(1 e )
x
Bn n
[ x an ]
n
n
!
n =1
= log x log a +
and hence we obtain
32
Bn n
[ x an ]
n =1 n n !
(1 e x )
(1 e a ) Bn n
= log
+
[ x an ]
x
a
n =1 n n !
(1 e a )
=0
a
x
B1 B2
n +1 Bn x
,
,...,
(
1)
Y
=
2
e x 1 n=0 1
n n!
n
Bn n B1 B2
n +1 Bn x
x
Y
,
,...,
(
1)
=
2
n n!
1
n =0 n !
n =0
B
B
B
Bn = Yn 1 , 2 ,..., (1) n +1 n
2
n
1
r
r B B
Br +1 = (1) k r k k +1
k =0
k k +1
33
This section is based on a paper by Snowden [22]. We have the well-known infinite
product
x2
sin x = x 1 2
k
k =1
and we have for x < 1
x2
(2n) 2 n
sin x
x
=
log
log
1
2
n
x k =1
n =1
k
bn n
x
n =1 n
= 2
where b2 n = 2 (2n) , b2 n +1 = 0 .
Therefore we have
sin x
xn
= Yn ( b1 , 1!b2 ,..., (n 1)!bn )
n!
x
n=0
2
3
= Y2 ( 0, 2 (2) ) or (2) =
2
6
4
60
= 2 (2) (4)
34
Y2 n =
Y2 n 1 k xk +1
k
k =0
n
n
2n 1
2n 1
Y
x
=
+
2 n 1 2 k 2 k +1
Y2 n 2 k x2 k
k = 0 2k
k =1 2k 1
This results in
n
2n 1 (1) n k (2n 2k )! 2 n 2 k (2k 1)! (2k )
(1) n (2n)! 2 n
= 2
(2n + 1)!
(2n 2k + 1)!
k =1 2k 1
which simplifies to
n
(2n)!
(1) k +1 2 k (2k )
= 2
(2n 1)!(2n + 1)!
(2n 2k + 1)!
k =1
(2k ) =
(1) k +1 22 k 2 k B2 k
2(2k )!
we obtain
n
22 k B2 k
(2n)!
=
(2n 1)!(2n + 1)! k =1 (2n 2k + 1)!(2k )!
or equivalently
n
2n 22 k B2 k
[(2n)!]2
=
(2n 1)!(2n + 1)! k =1 2k (2n 2k + 1)
REFERENCES
36
37