这是一篇随笔。前段时间数分课讲到了Taylor公式,做题的时候常常会用到 sin x , cos x \sin x,\cos x sinx,cosx等函数的Taylor展开式。我们容易写出 sin x \sin x sinx和 cos x \cos x cosx的 n n n阶麦克劳林展开式,但是 tan x \tan x tanx的麦克劳林展开式比较复杂,于是我翻开《具体数学》重新学习了一下跟伯努利数有关的内容,并做一些简单的整理。
首先给出伯努利数的定义。伯努利数
B
n
B_n
Bn是由如下递归式定义的数:
∑
j
=
0
m
(
m
+
1
j
)
B
j
=
[
m
=
0
]
,
m
∈
N
.
\sum\limits_{j=0}^m\binom{m+1}jB_j=[m=0],m\in\N.
j=0∑m(jm+1)Bj=[m=0],m∈N.
特别地,令
m
=
0
m=0
m=0可以得到
B
0
=
1
B_0=1
B0=1。
对这个递归式稍加改造,我们就得到了形式更整齐的
∑
k
=
0
n
(
n
k
)
B
k
=
B
n
+
[
n
=
1
]
,
n
∈
N
.
(1)
\sum\limits_{k=0}^n\binom n kB_k=B_n+[n=1],n\in\N.\tag1
k=0∑n(kn)Bk=Bn+[n=1],n∈N.(1)
假设
B
(
z
)
=
∑
n
⩾
0
B
n
z
n
n
!
\mathscr B(z)=\sum\limits_{n\geqslant0}B_n\dfrac{z^n}{n!}
B(z)=n⩾0∑Bnn!zn是伯努利数的指数生成函数,因为
e
z
=
∑
n
⩾
0
z
n
n
!
,
e^z=\sum\limits_{n\geqslant0}\dfrac{z^n}{n!},
ez=n⩾0∑n!zn,
所以它与
B
(
z
)
\mathscr B(z)
B(z)的卷积是
e
z
B
(
z
)
=
∑
n
⩾
0
∑
0
⩽
k
⩽
n
B
k
n
!
k
!
(
n
−
k
)
!
z
n
n
!
=
∑
n
⩾
0
(
∑
k
=
0
n
(
n
k
)
B
k
)
z
n
n
!
,
e^z\mathscr B(z)=\sum\limits_{n\geqslant0}\sum\limits_{0\leqslant k\leqslant n}\dfrac{B_kn!}{k!(n-k)!}\dfrac{z^n}{n!}=\sum\limits_{n\geqslant 0}\left(\sum\limits_{k=0}^n\binom n kB_k\right)\dfrac{z^n}{n!},
ezB(z)=n⩾0∑0⩽k⩽n∑k!(n−k)!Bkn!n!zn=n⩾0∑(k=0∑n(kn)Bk)n!zn,
而
1
1
1式的右边
B
n
+
[
n
=
1
]
B_n+[n=1]
Bn+[n=1]的指数生成函数是
B
(
z
)
+
z
\mathscr B(z)+z
B(z)+z,所以我们得到了
B
(
z
)
=
z
e
z
−
1
.
\mathscr B(z)=\dfrac{z}{e^z-1}.
B(z)=ez−1z.
注意到双曲函数
sinh
z
=
e
z
−
e
−
z
2
,
cosh
=
e
z
+
e
−
z
2
\sinh z=\dfrac{e^z-e^{-z}}{2},\cosh=\dfrac{e^z+e^{-z}}2
sinhz=2ez−e−z,cosh=2ez+e−z,则双曲余切函数
coth
z
=
e
z
+
e
−
z
e
z
−
e
−
z
\coth z=\dfrac{e^z+e^{-z}}{e^z-e^{-z}}
cothz=ez−e−zez+e−z,那么
z
2
coth
z
2
=
z
2
e
z
/
2
+
e
−
z
/
2
e
z
/
2
−
e
−
z
/
2
=
z
2
e
z
+
1
e
z
−
1
=
z
e
z
−
1
+
z
2
.
\dfrac{z}{2}\coth\dfrac z 2=\dfrac z 2\dfrac{e^{z/2}+e^{-z/2}}{e^{z/2}-e^{-z/2}}=\dfrac z 2\dfrac{e^z+1}{e^z-1}=\dfrac z{e^z-1}+\dfrac z 2.
2zcoth2z=2zez/2−e−z/2ez/2+e−z/2=2zez−1ez+1=ez−1z+2z.
又注意到
B
1
=
−
1
2
B_1=-\dfrac12
B1=−21,所以
z
e
z
−
1
+
z
2
=
∑
n
⩾
0
,
n
≠
1
B
n
z
n
n
!
\dfrac{z}{e^z-1}+\dfrac z2=\sum\limits_{n\geqslant 0,n\neq1}B_n\dfrac{z^n}{n!}
ez−1z+2z=n⩾0,n=1∑Bnn!zn。又因为
z
2
coth
z
2
\dfrac z2\coth\dfrac z2
2zcoth2z是奇函数,所以对所有大于
1
1
1的奇数
n
n
n都有
B
n
=
0
B_n=0
Bn=0,于是
z
e
z
−
1
+
z
2
=
∑
n
⩾
0
,
n
≠
1
B
n
z
n
n
!
\dfrac{z}{e^z-1}+\dfrac z 2=\sum\limits_{n\geqslant0,n\neq1}B_n\dfrac{z^n}{n!}
ez−1z+2z=n⩾0,n=1∑Bnn!zn中就只包含
n
n
n为偶数的项,我们可以把它写成
z
2
coth
z
2
=
∑
n
⩾
0
B
2
n
z
2
n
(
2
n
)
!
,
\dfrac z2\coth\dfrac z2=\sum\limits_{n\geqslant0}B_{2n}\dfrac{z^{2n}}{(2n)!},
2zcoth2z=n⩾0∑B2n(2n)!z2n,
用
2
z
2z
2z替换
z
z
z就有
z
coth
z
=
∑
n
⩾
0
B
2
n
4
n
z
2
n
(
2
n
)
!
.
z\coth z=\sum\limits_{n\geqslant 0}B_{2n}\dfrac{4^nz^{2n}}{(2n)!}.
zcothz=n⩾0∑B2n(2n)!4nz2n.
注意,这里的所有
z
z
z都是复数,而三角函数与双曲函数满足如下关系
sin
z
=
−
i
sinh
i
z
,
cos
z
=
cosh
i
z
,
cot
z
=
i
coth
i
z
,
\sin z=-i\sinh iz,\cos z=\cosh iz,\cot z=i\coth iz,
sinz=−isinhiz,cosz=coshiz,cotz=icothiz,
于是我们有
cot
z
=
i
z
coth
i
z
=
∑
n
⩾
0
B
2
n
(
−
4
)
n
z
2
n
−
1
(
2
n
)
!
.
\cot z=iz\coth iz=\sum\limits_{n\geqslant 0}B_{2n}\dfrac{(-4)^nz^{2n-1}}{(2n)!}.
cotz=izcothiz=n⩾0∑B2n(2n)!(−4)nz2n−1.
注意,
n
=
0
n=0
n=0时的求和项
B
2
n
(
−
4
)
n
z
2
n
−
1
(
2
n
)
!
=
1
z
B_{2n}\dfrac{(-4)^nz^{2n-1}}{(2n)!}=\dfrac1z
B2n(2n)!(−4)nz2n−1=z1不是
z
z
z的非负整数次幂,所以上式并不是关于
z
z
z的多项式。这是合理的,因为
cot
z
\cot z
cotz在
z
=
0
z=0
z=0处是发散到无穷大的,它不能在这一点处作麦克劳林展开。
从
cot
z
\cot z
cotz过渡到
tan
z
\tan z
tanz是非常容易的,我们并不需要做多项式除法,只要用
tan
z
=
cot
z
−
2
cot
2
z
\tan z=\cot z-2\cot 2z
tanz=cotz−2cot2z
即可得出
tan
z
=
∑
n
⩾
0
(
−
1
)
n
−
1
4
n
(
4
n
−
1
)
B
2
n
z
2
n
−
1
(
2
n
)
!
.
\tan z=\sum\limits_{n\geqslant 0}(-1)^{n-1}4^n(4^n-1)B_{2n}\dfrac{z^{2n-1}}{(2n)!}.
tanz=n⩾0∑(−1)n−14n(4n−1)B2n(2n)!z2n−1.
这是一个关于
z
z
z的多项式。虽然
n
=
0
n=0
n=0时
z
2
n
−
1
=
1
z
z^{2n-1}=\dfrac1z
z2n−1=z1,但这一项的系数含有
4
n
−
1
=
0
4^n-1=0
4n−1=0,因此它没有影响。
趁热打铁,做两个练习(《具体数学》习题6.72):求 z sin z \dfrac{z}{\sin z} sinzz和 ln tan z z \ln\dfrac{\tan z}z lnztanz的形式幂级数。
注意到
1
sin
z
+
cos
z
sin
z
=
1
+
cos
z
sin
z
=
cot
z
2
\dfrac{1}{\sin z}+\dfrac{\cos z}{\sin z}=\dfrac{1+\cos z}{\sin z}=\cot\dfrac z2
sinz1+sinzcosz=sinz1+cosz=cot2z,所以
z
sin
z
=
2
⋅
z
2
cot
z
2
−
z
cot
z
=
∑
n
⩾
0
(
−
1
)
n
(
2
−
4
n
)
B
2
n
z
2
n
(
2
n
)
!
.
\dfrac{z}{\sin z}=2\cdot\dfrac z2\cot\dfrac z2-z\cot z=\sum\limits_{n\geqslant0}(-1)^n(2-4^n)B_{2n}\dfrac{z^{2n}}{(2n)!}.
sinzz=2⋅2zcot2z−zcotz=n⩾0∑(−1)n(2−4n)B2n(2n)!z2n.
ln
tan
z
z
\ln\dfrac{\tan z}z
lnztanz可能有些棘手,但是注意到
2
sin
2
z
−
1
z
=
∑
n
⩾
1
(
−
4
)
n
(
2
−
4
n
)
B
2
n
z
2
n
−
1
(
2
n
)
!
,
\begin{aligned} \dfrac{2}{\sin 2z}-\dfrac1z=\sum\limits_{n\geqslant 1}(-4)^n(2-4^n)B_{2n}\dfrac{z^{2n-1}}{(2n)!}, \end{aligned}
sin2z2−z1=n⩾1∑(−4)n(2−4n)B2n(2n)!z2n−1,
两边关于
z
z
z取不定积分有
ln
tan
z
z
+
C
=
∑
n
⩾
1
(
−
4
)
n
(
2
−
4
n
)
B
2
n
2
n
z
2
n
(
2
n
)
!
,
\ln\dfrac{\tan z}z+C=\sum\limits_{n\geqslant 1}(-4)^n(2-4^n)\dfrac{B_{2n}}{2n}\dfrac{z^{2n}}{(2n)!},
lnztanz+C=n⩾1∑(−4)n(2−4n)2nB2n(2n)!z2n,
注意到
z
=
0
z=0
z=0时
ln
tan
z
z
=
0
\ln\dfrac{\tan z}z=0
lnztanz=0,所以
C
=
0
C=0
C=0。因此我们得到了
ln
tan
z
z
=
∑
n
⩾
1
(
−
4
)
n
(
2
−
4
n
)
B
2
n
2
n
⋅
z
2
n
(
2
n
)
!
.
\ln\dfrac{\tan z}z=\sum\limits_{n\geqslant1}(-4)^n(2-4^n)\dfrac{B_{2n}}{2n}\cdot\dfrac{z^{2n}}{(2n)!}.
lnztanz=n⩾1∑(−4)n(2−4n)2nB2n⋅(2n)!z2n.
关于伯努利数,它最初是由伯努利在自然数的幂之和的式子中发现的,即
S
m
(
n
)
=
∑
k
=
0
n
−
1
k
m
=
1
m
+
1
∑
k
=
0
m
(
m
+
1
k
)
B
k
n
m
+
1
−
k
,
m
∈
N
.
(2)
S_m(n)=\sum\limits_{k=0}^{n-1}k^m=\dfrac{1}{m+1}\sum\limits_{k=0}^m\binom{m+1}{k}B_kn^{m+1-k},m\in\N.\tag{2}
Sm(n)=k=0∑n−1km=m+11k=0∑m(km+1)Bknm+1−k,m∈N.(2)
也就是说,
S
m
(
n
)
S_m(n)
Sm(n)总是关于
n
n
n的
m
+
1
m+1
m+1次多项式,而这个多项式系数与伯努利数有关。这个结论可以利用数学归纳法证明。
**证 **首先, S 0 ( n ) = n , 1 0 + 1 ( 1 0 ) B 0 n 0 + 1 − 0 = n S_0(n)=n,\dfrac{1}{0+1}\dbinom{1}{0}B_0n^{0+1-0}=n S0(n)=n,0+11(01)B0n0+1−0=n,这时 2 2 2式成立。
现假设对所有
0
⩽
j
<
m
0\leqslant j<m
0⩽j<m,都有
S
j
(
n
)
=
1
j
+
1
∑
k
=
0
j
(
j
+
1
k
)
B
k
n
j
+
1
−
k
,
S_j(n)=\dfrac{1}{j+1}\sum\limits_{k=0}^j\binom{j+1}{k}B_kn^{j+1-k},
Sj(n)=j+11k=0∑j(kj+1)Bknj+1−k,
考虑求出
S
m
(
n
)
S_m(n)
Sm(n)。为此,我们使用扰动法:
S
m
+
1
(
n
)
+
n
m
+
1
=
∑
k
=
0
n
k
m
+
1
=
∑
k
=
0
n
−
1
(
k
+
1
)
m
+
1
=
∑
k
=
0
n
−
1
∑
j
=
0
m
+
1
(
m
+
1
j
)
k
j
=
∑
j
=
0
m
+
1
(
m
+
1
j
)
∑
k
=
0
n
−
1
k
j
=
∑
j
=
0
m
+
1
(
m
+
1
j
)
S
j
(
n
)
.
\begin{aligned} S_{m+1}(n)+n^{m+1} &=\sum\limits_{k=0}^nk^{m+1}\\ &=\sum\limits_{k=0}^{n-1}(k+1)^{m+1}\\ &=\sum\limits_{k=0}^{n-1}\sum\limits_{j=0}^{m+1}\binom{m+1}{j}k^j\\ &=\sum\limits_{j=0}^{m+1}\binom{m+1}{j}\sum\limits_{k=0}^{n-1}k^j\\ &=\sum\limits_{j=0}^{m+1}\binom{m+1}{j}S_j(n). \end{aligned}
Sm+1(n)+nm+1=k=0∑nkm+1=k=0∑n−1(k+1)m+1=k=0∑n−1j=0∑m+1(jm+1)kj=j=0∑m+1(jm+1)k=0∑n−1kj=j=0∑m+1(jm+1)Sj(n).
两边消去
S
m
+
1
(
n
)
S_{m+1}(n)
Sm+1(n)得
n
m
+
1
=
∑
j
=
0
m
(
m
+
1
j
)
S
j
(
n
)
,
n^{m+1}=\sum\limits_{j=0}^m\binom{m+1}{j}S_j(n),
nm+1=j=0∑m(jm+1)Sj(n),
于是
(
m
+
1
)
S
m
(
n
)
=
n
m
+
1
−
∑
j
=
0
m
−
1
(
m
+
1
j
)
1
j
+
1
∑
k
=
0
j
(
j
+
1
k
)
B
k
n
j
+
1
−
k
=
n
m
+
1
−
∑
j
=
0
m
−
1
(
m
+
1
j
)
1
j
+
1
∑
k
=
0
j
(
j
+
1
k
+
1
)
B
j
−
k
n
k
+
1
=
n
m
+
1
−
∑
k
=
0
m
−
1
n
k
+
1
∑
j
=
k
m
−
1
1
j
+
1
(
m
+
1
j
)
(
j
+
1
k
+
1
)
B
j
−
k
=
n
m
+
1
−
∑
k
=
0
m
−
1
n
k
+
1
∑
j
=
0
m
−
1
−
k
1
j
+
k
+
1
(
m
+
1
j
+
k
)
(
j
+
k
+
1
k
+
1
)
B
j
=
n
m
+
1
−
∑
k
=
0
m
−
1
n
k
+
1
∑
j
=
0
m
−
1
−
l
1
k
+
1
(
m
+
1
j
+
k
)
(
j
+
k
k
)
B
j
=
n
m
+
1
−
∑
k
=
0
m
−
1
n
k
+
1
k
+
1
∑
j
=
0
m
−
1
−
k
(
m
+
1
k
)
(
m
+
1
−
k
j
)
B
j
=
n
m
+
1
−
∑
k
=
0
m
−
1
n
k
+
1
k
+
1
(
m
+
1
k
)
∑
j
=
0
m
−
1
−
k
(
m
+
1
−
k
j
)
B
j
\begin{aligned} (m+1)S_m(n) &=n^{m+1}-\sum\limits_{j=0}^{m-1}\binom{m+1}{j}\dfrac{1}{j+1}\sum\limits_{k=0}^j\binom{j+1}{k}B_kn^{j+1-k}\\ &=n^{m+1}-\sum\limits_{j=0}^{m-1}\binom{m+1}{j}\dfrac{1}{j+1}\sum\limits_{k=0}^j\binom{j+1}{k+1}B_{j-k}n^{k+1}\\ &=n^{m+1}-\sum\limits_{k=0}^{m-1}n^{k+1}\sum\limits_{j=k}^{m-1}\dfrac{1}{j+1}\binom{m+1}{j}\binom{j+1}{k+1}B_{j-k}\\ &=n^{m+1}-\sum\limits_{k=0}^{m-1}n^{k+1}\sum\limits_{j=0}^{m-1-k}\dfrac{1}{j+k+1}\binom{m+1}{j+k}\binom{j+k+1}{k+1}B_j\\ &=n^{m+1}-\sum\limits_{k=0}^{m-1}n^{k+1}\sum\limits_{j=0}^{m-1-l}\dfrac{1}{k+1}\binom{m+1}{j+k}\binom{j+k}{k}B_j\\ &=n^{m+1}-\sum\limits_{k=0}^{m-1}\dfrac{n^{k+1}}{k+1}\sum\limits_{j=0}^{m-1-k}\binom{m+1}{k}\binom{m+1-k}{j}B_j\\ &=n^{m+1}-\sum\limits_{k=0}^{m-1}\dfrac{n^{k+1}}{k+1}\binom{m+1}{k}\sum\limits_{j=0}^{m-1-k}\binom{m+1-k}jB_j \end{aligned}
(m+1)Sm(n)=nm+1−j=0∑m−1(jm+1)j+11k=0∑j(kj+1)Bknj+1−k=nm+1−j=0∑m−1(jm+1)j+11k=0∑j(k+1j+1)Bj−knk+1=nm+1−k=0∑m−1nk+1j=k∑m−1j+11(jm+1)(k+1j+1)Bj−k=nm+1−k=0∑m−1nk+1j=0∑m−1−kj+k+11(j+km+1)(k+1j+k+1)Bj=nm+1−k=0∑m−1nk+1j=0∑m−1−lk+11(j+km+1)(kj+k)Bj=nm+1−k=0∑m−1k+1nk+1j=0∑m−1−k(km+1)(jm+1−k)Bj=nm+1−k=0∑m−1k+1nk+1(km+1)j=0∑m−1−k(jm+1−k)Bj
因为这里的
k
<
m
k<m
k<m,那么
∑
j
=
0
m
−
k
(
m
+
1
−
k
j
)
B
j
=
[
m
=
k
]
=
0
,
\sum\limits_{j=0}^{m-k}\binom{m+1-k}jB_j=[m=k]=0,
j=0∑m−k(jm+1−k)Bj=[m=k]=0,
所以
∑
j
=
0
m
−
1
−
k
(
m
+
1
−
k
j
)
B
j
=
−
(
m
+
1
−
k
)
B
m
−
k
,
\sum\limits_{j=0}^{m-1-k}\binom{m+1-k}jB_j=-(m+1-k)B_{m-k},
j=0∑m−1−k(jm+1−k)Bj=−(m+1−k)Bm−k,
那么
(
m
+
1
)
S
m
(
n
)
=
n
m
+
1
+
∑
k
=
0
m
−
1
n
k
+
1
k
+
1
(
m
+
1
k
)
(
m
+
1
−
k
)
B
m
−
k
=
∑
k
=
0
m
n
k
+
1
k
+
1
(
m
+
1
k
)
(
m
+
1
−
k
)
B
m
−
k
⇒
S
m
(
n
)
=
1
m
+
1
∑
k
=
0
m
(
m
+
1
k
)
B
k
n
m
−
k
+
1
.
\begin{aligned} (m+1)S_m(n) &=n^{m+1}+\sum\limits_{k=0}^{m-1}\dfrac{n^{k+1}}{k+1}\binom{m+1}{k}(m+1-k)B_{m-k}\\ &=\sum\limits_{k=0}^m\dfrac{n^{k+1}}{k+1}\binom{m+1}{k}(m+1-k)B_{m-k}\\ \Rightarrow S_m(n)&=\dfrac{1}{m+1}\sum\limits_{k=0}^m\binom{m+1}{k}B_kn^{m-k+1}. \end{aligned}
(m+1)Sm(n)⇒Sm(n)=nm+1+k=0∑m−1k+1nk+1(km+1)(m+1−k)Bm−k=k=0∑mk+1nk+1(km+1)(m+1−k)Bm−k=m+11k=0∑m(km+1)Bknm−k+1.
由第二数学归纳法,
2
2
2式对任意自然数
m
m
m成立。
□
\Box
□
伯努利数还有其他应用,其中尤为重要的一个是它可以帮我们求出偶数阶调和数。首先有
z
cot
z
=
1
−
2
∑
k
⩾
1
z
2
k
2
π
2
−
z
2
.
(3)
z\cot z=1-2\sum\limits_{k\geqslant 1}\dfrac{z^2}{k^2\pi^2-z^2}.\tag3
zcotz=1−2k⩾1∑k2π2−z2z2.(3)
这个式子的证明放到后面。现在将
z
2
k
2
π
2
−
z
2
=
z
2
k
2
π
2
1
−
z
2
k
2
π
2
=
−
1
+
1
1
−
z
2
k
2
π
2
\dfrac{z^2}{k^2\pi^2-z^2}=\dfrac{\frac{z^2}{k^2\pi^2}}{1-\frac{z^2}{k^2\pi^2}}=-1+\dfrac{1}{1-\frac{z^2}{k^2\pi^2}}
k2π2−z2z2=1−k2π2z2k2π2z2=−1+1−k2π2z21展开为形式幂级数,有
z
2
k
2
π
2
−
z
2
=
∑
n
⩾
1
(
z
2
k
2
π
2
)
n
,
\dfrac{z^2}{k^2\pi^2-z^2}=\sum\limits_{n\geqslant1}\left(\dfrac{z^2}{k^2\pi^2}\right)^n,
k2π2−z2z2=n⩾1∑(k2π2z2)n,
于是
z
cot
z
=
1
−
2
∑
k
⩾
1
∑
n
⩾
1
z
2
n
π
2
n
⋅
1
k
2
n
=
1
−
2
∑
n
⩾
1
z
2
n
π
2
n
∑
k
⩾
1
1
k
2
n
=
1
−
2
∑
n
⩾
1
z
2
n
π
2
n
H
∞
(
2
n
)
.
z\cot z=1-2\sum\limits_{k\geqslant 1}\sum\limits_{n\geqslant 1}\dfrac{z^{2n}}{\pi^{2n}}\cdot\dfrac{1}{k^{2n}}=1-2\sum\limits_{n\geqslant 1}\dfrac{z^{2n}}{\pi^{2n}}\sum\limits_{k\geqslant 1}\dfrac{1}{k^{2n}}=1-2\sum\limits_{n\geqslant 1}\dfrac{z^{2n}}{\pi^{2n}}H_\infty^{(2n)}.
zcotz=1−2k⩾1∑n⩾1∑π2nz2n⋅k2n1=1−2n⩾1∑π2nz2nk⩾1∑k2n1=1−2n⩾1∑π2nz2nH∞(2n).
又因为
z
cot
z
=
∑
n
⩾
0
(
−
4
)
n
B
2
n
z
2
n
(
2
n
)
!
,
z\cot z=\sum\limits_{n\geqslant 0}(-4)^nB_{2n}\dfrac{z^{2n}}{(2n)!},
zcotz=n⩾0∑(−4)nB2n(2n)!z2n,
取含有
z
2
n
z^{2n}
z2n的项的系数,就得到
H
∞
(
2
n
)
=
(
−
1
)
n
−
1
2
2
n
−
1
B
2
n
π
2
n
(
2
n
)
!
.
H_\infty^{(2n)}=\dfrac{(-1)^{n-1}2^{2n-1}B_{2n}\pi^{2n}}{(2n)!}.
H∞(2n)=(2n)!(−1)n−122n−1B2nπ2n.
例如我们熟知的
H
∞
(
2
)
=
∑
n
⩾
1
1
n
2
=
π
2
6
,
H
∞
(
4
)
=
∑
n
⩾
1
1
n
4
=
π
4
90
H_\infty^{(2)}=\sum\limits_{n\geqslant 1}\dfrac{1}{n^2}=\dfrac{\pi^2}{6},\\ H_\infty^{(4)}=\sum\limits_{n\geqslant 1}\dfrac{1}{n^4}=\dfrac{\pi^4}{90}
H∞(2)=n⩾1∑n21=6π2,H∞(4)=n⩾1∑n41=90π4
等等。(Rmk:这里的
H
∞
(
2
n
)
H_\infty^{(2n)}
H∞(2n)也可以记作黎曼
ζ
\zeta
ζ函数的形式:
ζ
(
2
n
)
\zeta(2n)
ζ(2n))
这里补上
3
3
3式的证明。这个式子的证明来源于《具体数学》的习题6.73,首先证明
z
cot
z
=
z
2
n
−
1
cot
z
2
n
−
1
+
∑
k
=
1
2
n
−
1
−
1
z
2
n
(
cot
z
+
k
π
2
n
+
cot
z
−
k
π
2
n
)
.
(4)
z\cot z=\dfrac{z}{2^{n-1}}\cot\dfrac{z}{2^{n-1}}+\sum\limits_{k=1}^{2^{n-1}-1}\dfrac{z}{2^n}\left(\cot\dfrac{z+k\pi}{2^n}+\cot\dfrac{z-k\pi}{2^n}\right).\tag4
zcotz=2n−1zcot2n−1z+k=1∑2n−1−12nz(cot2nz+kπ+cot2nz−kπ).(4)
我们有
z
2
n
−
1
cot
z
2
n
−
1
+
∑
k
=
1
2
n
−
1
−
1
z
2
n
(
cot
z
+
k
π
2
n
+
cot
z
−
k
π
2
n
)
=
z
2
n
−
1
cot
z
2
n
−
1
+
z
2
n
(
∑
k
=
1
2
n
−
1
−
1
cot
z
+
k
π
2
n
+
∑
k
=
1
2
n
−
1
−
1
cot
z
−
(
2
n
−
1
−
k
)
π
2
n
)
=
z
2
n
−
1
cot
z
2
n
−
1
+
z
2
n
(
∑
k
=
1
2
n
−
1
−
1
cot
z
+
k
π
2
n
−
∑
k
=
1
2
n
−
1
−
1
tan
z
+
k
π
2
n
)
=
z
2
n
−
1
cot
z
2
n
−
1
+
z
2
n
−
1
∑
k
=
1
2
n
−
1
−
1
cot
z
+
k
π
2
n
−
1
=
z
2
n
−
1
∑
k
=
1
2
n
−
1
−
1
cot
z
+
k
π
2
n
−
1
\begin{aligned} &\dfrac{z}{2^{n-1}}\cot\dfrac{z}{2^{n-1}}+\sum\limits_{k=1}^{2^{n-1}-1}\dfrac{z}{2^n}\left(\cot\dfrac{z+k\pi}{2^n}+\cot\dfrac{z-k\pi}{2^n}\right)\\ &=\dfrac{z}{2^{n-1}}\cot\dfrac{z}{2^{n-1}}+\dfrac{z}{2^n}\left(\sum\limits_{k=1}^{2^{n-1}-1}\cot\dfrac{z+k\pi}{2^n}+\sum\limits_{k=1}^{2^{n-1}-1}\cot\dfrac{z-(2^{n-1}-k)\pi}{2^n}\right)\\ &=\dfrac{z}{2^{n-1}}\cot\dfrac{z}{2^{n-1}}+\dfrac{z}{2^n}\left(\sum\limits_{k=1}^{2^{n-1}-1}\cot\dfrac{z+k\pi}{2^n}-\sum\limits_{k=1}^{2^{n-1}-1}\tan\dfrac{z+k\pi}{2^n}\right)\\ &=\dfrac{z}{2^{n-1}}\cot\dfrac{z}{2^{n-1}}+\dfrac{z}{2^{n-1}}\sum\limits_{k=1}^{2^{n-1}-1}\cot\dfrac{z+k\pi}{2^{n-1}}\\ &=\dfrac{z}{2^{n-1}}\sum\limits_{k=1}^{2^{n-1}-1}\cot\dfrac{z+k\pi}{2^{n-1}} \end{aligned}
2n−1zcot2n−1z+k=1∑2n−1−12nz(cot2nz+kπ+cot2nz−kπ)=2n−1zcot2n−1z+2nz⎝⎛k=1∑2n−1−1cot2nz+kπ+k=1∑2n−1−1cot2nz−(2n−1−k)π⎠⎞=2n−1zcot2n−1z+2nz⎝⎛k=1∑2n−1−1cot2nz+kπ−k=1∑2n−1−1tan2nz+kπ⎠⎞=2n−1zcot2n−1z+2n−1zk=1∑2n−1−1cot2n−1z+kπ=2n−1zk=1∑2n−1−1cot2n−1z+kπ
再利用数学归纳法(略)即可证明上式等于
z
cot
z
z\cot z
zcotz。
下面对
4
4
4式中的第
k
k
k个求和项取
n
→
∞
n\to\infty
n→∞时的极限,有
lim
n
→
∞
z
2
n
(
cot
z
+
k
π
2
n
+
cot
z
−
k
π
2
n
)
=
lim
n
→
∞
z
2
n
sin
z
2
n
−
1
sin
z
+
k
π
2
n
sin
z
−
k
π
2
n
=
lim
n
→
∞
z
2
n
z
2
n
−
1
z
+
k
π
2
n
z
−
k
π
2
n
=
2
z
2
z
2
−
k
2
π
2
\begin{aligned} &\lim\limits_{n\to\infty}\dfrac{z}{2^n}\left(\cot\dfrac{z+k\pi}{2^n}+\cot\dfrac{z-k\pi}{2^n}\right)\\ &=\lim\limits_{n\to\infty}\dfrac{z}{2^n}\dfrac{\sin\frac{z}{2^{n-1}}}{\sin\frac{z+k\pi}{2^n}\sin\frac{z-k\pi}{2^n}}\\ &=\lim\limits_{n\to\infty}\dfrac{z}{2^n}\dfrac{\frac{z}{2^{n-1}}}{\frac{z+k\pi}{2^n}\frac{z-k\pi}{2^n}}\\ &=\dfrac{2z^2}{z^2-k^2\pi^2} \end{aligned}
n→∞lim2nz(cot2nz+kπ+cot2nz−kπ)=n→∞lim2nzsin2nz+kπsin2nz−kπsin2n−1z=n→∞lim2nz2nz+kπ2nz−kπ2n−1z=z2−k2π22z2
又因为
lim
n
→
∞
z
2
n
−
1
cot
z
2
n
−
1
=
1
\lim\limits_{n\to\infty}\dfrac{z}{2^{n-1}}\cot\dfrac{z}{2^{n-1}}=1
n→∞lim2n−1zcot2n−1z=1,所以
z
cot
z
=
1
−
2
∑
k
⩾
1
z
2
k
2
π
2
−
z
2
.
□
z\cot z=1-2\sum\limits_{k\geqslant 1}\dfrac{z^2}{k^2\pi^2-z^2}.\Box
zcotz=1−2k⩾1∑k2π2−z2z2.□