本篇是黎曼 ζ \zeta ζ函数系列的第二篇,传送门在此书接上回,让我们继续出发。
著名的欧拉乘积公式
ζ ( s ) = ∑ n = 1 ∞ 1 n s = ∏ p 1 1 − 1 p s \zeta(s) = \sum_{n = 1}^\infin \frac{1}{n^s} = \prod_p\frac{1}{1 - \frac{1}{p^s}} ζ(s)=n=1∑∞ns1=p∏1−ps11
揭开了素数分布秘密的一角。证明思路如下,由
ζ ( s ) = 1 1 s + 1 2 s + 1 3 s + ⋯ \zeta(s) = \frac{1}{1^s} + \frac{1}{2^s} + \frac{1}{3^s} + \cdots ζ(s)=1s1+2s1+3s1+⋯
两边乘以 1 2 s \frac{1}{2^s} 2s1就得到
1 2 s ζ ( s ) = 1 2 s + 1 4 s + 1 6 s + ⋯ \frac{1}{2^s}\zeta(s) = \frac{1}{2^s} + \frac{1}{4^s} + \frac{1}{6^s} + \cdots 2s1ζ(s)=2s1+4s1+6s1+⋯
两式相减得(减掉所有2的倍数项)
( 1 − 1 2 s ) ζ ( s ) = 1 1 s + 1 3 s + 1 5 s + ⋯ (1 - \frac{1}{2^s})\zeta(s) = \frac{1}{1^s} + \frac{1}{3^s} + \frac{1}{5^s} + \cdots (1−2s1)ζ(s)=1s1+3s1+5s1+⋯
两边乘以 1 3 s \frac{1}{3^s} 3s1就得到
1 3 s ( 1 − 1 2 s ) ζ ( s ) = 1 3 s + 1 9 s + 1 1 5 s + ⋯ \frac{1}{3^s}(1 - \frac{1}{2^s})\zeta(s) = \frac{1}{3^s} + \frac{1}{9^s} + \frac{1}{15^s} + \cdots 3s1(1−2s1)ζ(s)=3s1+9s1+15s1+⋯
再相减就得到(减掉所有3的倍数项)
( 1 − 1 3 s ) ( 1 − 1 2 s ) ζ ( s ) = 1 1 s + 1 5 s + 1 7 s + ⋯ (1 - \frac{1}{3^s})(1 - \frac{1}{2^s})\zeta(s) = \frac{1}{1^s} + \frac{1}{5^s} + \frac{1}{7^s} + \cdots (1−3s1)(1−2s1)ζ(s)=1s1+5s1+7s1+⋯
两边乘以 1 5 s \frac{1}{5^s} 5s1再相减(减掉所有5的倍数项),如此这般下去,干掉右边所有得项,最终得到
⋯ ( 1 − 1 5 s ) ( 1 − 1 3 s ) ( 1 − 1 2 s ) ζ ( s ) = 1 \cdots(1 - \frac{1}{5^s})(1 - \frac{1}{3^s})(1 - \frac{1}{2^s})\zeta(s) = 1 ⋯(1−5s1)(1−3s1)(1−2s1)ζ(s)=1
不难发现,乘到左边的项都是素数项,所以就有
ζ ( s ) = ∏ p 1 1 − 1 p s \zeta(s) = \prod_p\frac{1}{1 - \frac{1}{p^s}} ζ(s)=p∏1−ps11
让我们尝试展开上式的右侧,得到
1 ζ ( s ) = ( 1 − 1 2 s ) ( 1 − 1 3 s ) ( 1 − 1 5 s ) ( 1 − 1 7 s ) ⋯ = 1 − 1 2 s − 1 3 s − 1 5 s + 1 6 s − 1 7 s + ⋯ \frac{1}{\zeta(s)} = (1 - \frac{1}{2^s})(1 - \frac{1}{3^s})(1 - \frac{1}{5^s})(1 - \frac{1}{7^s})\cdots \\ = 1 - \frac{1}{2^s} - \frac{1}{3^s} - \frac{1}{5^s} + \frac{1}{6^s} - \frac{1}{7^s} + \cdots ζ(s)1=(1−2s1)(1−3s1)(1−5s1)(1−7s1)⋯=1−2s1−3s1−5s1+6s1−7s1+⋯
如果把这个级数写成Dirichlet级数
1 ζ ( s ) = ∑ n a n n s \frac{1}{\zeta(s)} = \sum_n\frac{a_n}{n^s} ζ(s)1=n∑nsan
的形式,就会有当 n n n为偶数个不同素数乘积时 a n = 1 a_n = 1 an=1,当 n n n为奇数个不同素数乘积时 a n = − 1 a_n = -1 an=−1,而当 n n n可被某一素数的平方整除时 a n = 0 a_n = 0 an=0. 我们就定义这样的 a n a_n an为Möbius函数,记为 μ ( n ) \mu(n) μ(n),于是
1 ζ ( s ) = ∑ n μ ( n ) n s \frac{1}{\zeta(s)} = \sum_n\frac{\mu(n)}{n^s} ζ(s)1=n∑nsμ(n)
容易验证Möbius函数是积性函数,也就是说 μ ( 1 ) = 1 \mu(1) = 1 μ(1)=1,且当 a , b a, ~ b a, b互质时, μ ( a b ) = μ ( a ) μ ( b ) \mu(ab) = \mu(a)\mu(b) μ(ab)=μ(a)μ(b).
欧拉注意到,当 s = 1 s = 1 s=1时, ζ ( 1 ) \zeta(1) ζ(1)为调和级数,其以对数方式发散。实际上
ln ( x ) = ∫ 1 x 1 t d t \ln(x) = \int_1^x\frac{1}{t}dt ln(x)=∫1xt1dt
为了干掉连乘积,两边取对数就有
ln ( ∑ n 1 n ) = − ∑ p ln ( 1 − 1 p ) \ln(\sum_n\frac{1}{n}) = -\sum_p\ln(1 - \frac{1}{p}) ln(n∑n1)=−p∑ln(1−p1)
为了展开 ln ( 1 − 1 p ) \ln(1 - \frac{1}{p}) ln(1−p1)为幂级数,我们记 q = 1 p q = \frac{1}{p} q=p1,显然 ln ( 1 − q ) \ln(1 - q) ln(1−q)可以在 q = 0 q = 0 q=0处展开,因为其 n n n阶导数在 q = 0 q = 0 q=0处为 − ( n − 1 ) ! -(n - 1)! −(n−1)!,于是
− ln ( 1 − q ) = q + 1 2 q 2 + 1 3 q 3 + ⋯ -\ln(1 - q) = q + \frac{1}{2}q^2 + \frac{1}{3}q^3 + \cdots −ln(1−q)=q+21q2+31q3+⋯
代入原式就得到
ln ( ∑ n 1 n ) = − ∑ p ln ( 1 − 1 p ) = ∑ p ( 1 p + 1 2 p 2 + 1 3 p 3 + ⋯ ) \ln(\sum_n\frac{1}{n}) = -\sum_p\ln(1 - \frac{1}{p}) = \sum_p(\frac{1}{p} + \frac{1}{2p^2} + \frac{1}{3p^3} + \cdots) ln(n∑n1)=−p∑ln(1−p1)=p∑(p1+2p21+3p31+⋯)
上式右边除第一项外是收敛的,为了看出这一点,只需注意到
∑ k = 2 ∞ ( 1 2 k 2 + 1 3 k 3 + ⋯ ) = − ∑ k = 2 ∞ [ 1 k + ln ( 1 − 1 k ) ] = 1 − ∑ k = 1 ∞ 1 k + ln ∏ k = 2 ∞ k k − 1 = 1 − lim n → ∞ [ ∑ k = 1 n 1 k − ln ( n ) ] = 1 − γ \sum_{k = 2}^\infty(\frac{1}{2k^2} + \frac{1}{3k^3} + \cdots) = -\sum_{k = 2}^\infty[\frac{1}{k} + \ln(1 - \frac{1}{k})] = 1 - \sum_{k = 1}^\infty\frac{1}{k} + \ln\prod_{k = 2}^\infty\frac{k}{k - 1} = 1 - \lim_{n \to \infty}[\sum_{k = 1}^n\frac{1}{k} - \ln(n)] = 1 - \gamma k=2∑∞(2k21+3k31+⋯)=−k=2∑∞[k1+ln(1−k1)]=1−k=1∑∞k1+lnk=2∏∞k−1k=1−n→∞lim[k=1∑nk1−ln(n)]=1−γ
其中 γ \gamma γ为Euler-Mascheroni常数。于是
∑ p 1 p ∼ ln ( ∑ n 1 n ) ∼ ln ln ( n ) \sum_p\frac{1}{p} \sim \ln(\sum_n\frac{1}{n}) \sim \ln\ln(n) p∑p1∼ln(n∑n1)∼lnln(n)
或者更确切地说
∑ p < N 1 p ∼ ln ln ( N ) \sum_{p < N}\frac{1}{p} \sim \ln\ln(N) p<N∑p1∼lnln(N)
更进一步地,上式左边似乎可以改写为积分的形式 ∫ N 1 x d x \int^N\frac{1}{x}dx ∫Nx1dx,但实际上并非所有的 x x x都是素数,我们需要修改这个积分为
∫ N 1 x ρ ( x ) d x \int^N\frac{1}{x}\rho(x)dx ∫Nx1ρ(x)dx
其中 ρ ( x ) \rho(x) ρ(x)给出在 x x x附近单位区间内存在素数的概率。如果注意到 ln ln ( N ) = ∫ N 1 x ln ( x ) d x \ln\ln(N) = \int^N\frac{1}{x\ln(x)}dx lnln(N)=∫Nxln(x)1dx,就有
ρ ( x ) ∼ 1 ln ( x ) \rho(x) \sim \frac{1}{\ln(x)} ρ(x)∼ln(x)1
于是 x x x以内的素数个数 π ( x ) \pi(x) π(x)可以表示为
π ( x ) = ∫ ρ ( x ) d x ∼ ∫ 1 ln ( x ) d x ∼ x ln ( x ) \pi(x) = \int\rho(x)dx \sim \int\frac{1}{\ln(x)}dx \sim \frac{x}{\ln(x)} π(x)=∫ρ(x)dx∼∫ln(x)1dx∼ln(x)x
这就是素数定理。
函数 f : Z + → C f \colon \Z^+ \to \mathbb C f:Z+→C被称为数论函数,每一个数论函数均可被视为一个复数序列。
对于数论函数 f , g f, ~ g f, g,其Dirichlet卷积定义为
( f ∗ g ) ( n ) = ∑ d ∣ n f ( d ) g ( n d ) (f * g)(n) = \sum_{d \mid n}f(d)g(\frac{n}{d}) (f∗g)(n)=d∣n∑f(d)g(dn)
可以验证,所有数论函数的集合以Dirichlet卷积为乘法构成一个阿贝尔幺半群。
实际上这个阿贝尔幺半群是一个整环(无零因子阿贝尔幺环),也就是说
- 数论函数对普通加法构成阿贝尔群
- 非零数论函数对卷积构成阿贝尔幺半群(即无零因子阿贝尔幺半群)
- 卷积对加法有分配律
以下验证相关性质,首先,根据定义,如果我们记 b ≡ n d b \equiv \frac{n}{d} b≡dn,就容易看出
( f ∗ g ) ( n ) = ∑ d ∣ n f ( d ) g ( n d ) = ∑ b ∣ n g ( b ) f ( n b ) = ( g ∗ f ) ( n ) (f * g)(n) = \sum_{d \mid n}f(d)g(\frac{n}{d}) = \sum_{b \mid n}g(b)f(\frac{n}{b}) = (g*f)(n) (f∗g)(n)=d∣n∑f(d)g(dn)=b∣n∑g(b)f(bn)=(g∗f)(n)
而注意到这个交换律我们就有
( ( f ∗ g ) ∗ h ) ( n ) = ∑ m ∣ n [ ∑ d ∣ m f ( d ) g ( m d ) ] h ( n m ) = ∑ m ∣ n h ( m ) ∑ d ∣ n m f ( d ) g ( n d m ) = ∑ m d = n h ( m ) f ( d ) g ( n d m ) = ∑ d ∣ n f ( d ) ∑ m ∣ n d h ( m ) g ( n d m ) ((f * g) * h)(n) = \sum_{m \mid n}[\sum_{d \mid m}f(d)g(\frac{m}{d})]h(\frac{n}{m}) = \sum_{m \mid n}h(m)\sum_{d \mid \frac{n}{m}}f(d)g(\frac{n}{dm}) \\ = \sum_{md = n}h(m)f(d)g(\frac{n}{dm}) = \sum_{d \mid n}f(d)\sum_{m \mid \frac{n}{d}}h(m)g(\frac{n}{dm}) ((f∗g)∗h)(n)=m∣n∑[d∣m∑f(d)g(dm)]h(mn)=m∣n∑h(m)d∣mn∑f(d)g(dmn)=md=n∑h(m)f(d)g(dmn)=d∣n∑f(d)m∣dn∑h(m)g(dmn)
而
( f ∗ ( g ∗ h ) ) ( n ) = ∑ d ∣ n f ( d ) ∑ m ∣ n d g ( m ) h ( n d m ) = ∑ d ∣ n f ( d ) ∑ m ∣ n d h ( m ) g ( n d m ) (f * (g * h))(n) = \sum_{d \mid n}f(d)\sum_{m \mid \frac{n}{d}}g(m)h(\frac{n}{dm}) = \sum_{d \mid n}f(d)\sum_{m \mid \frac{n}{d}}h(m)g(\frac{n}{dm}) (f∗(g∗h))(n)=d∣n∑f(d)m∣dn∑g(m)h(dmn)=d∣n∑f(d)m∣dn∑h(m)g(dmn)
这就证明了结合律。
证明
( 1 ∗ μ ) ( 1 ) = 1 (1 * \mu)(1) = 1 (1∗μ)(1)=1是显然的。当 n > 1 n > 1 n>1时,根据算数基本定理,我们可将 n n n唯一分解为素数幂的乘积。 n = ∏ k = 1 m p k a k n = \prod^m_{k = 1}p_k^{a_k} n=∏k=1mpkak。现在设 d d d的分解为 d = ∏ k = 1 m p k b k , 0 ≤ b k ≤ a k d = \prod^m_{k = 1}p_k^{b_k}, ~ 0 \le b_k \le a_k d=∏k=1mpkbk, 0≤bk≤ak,注意到如果 b k ≥ 2 b_k \ge 2 bk≥2就有 μ ( d ) = 0 \mu(d) = 0 μ(d)=0,因此
( 1 ∗ μ ) ( n ) = ∑ b k = 0 , 1 μ ( p 1 b 1 p 2 b 2 ⋯ p m b m ) = ∑ b k = 0 , 1 μ ( p 1 b 1 ) μ ( p 2 b 2 ) ⋯ μ ( p m b m ) = ∑ b k = 0 , 1 ( − 1 ) b 1 + b 2 + ⋯ + b m = ∑ s m ( m s ) ( − 1 ) s = ( 1 − 1 ) m = 0 \begin{aligned} (1 * \mu)(n) &= \sum_{b_k = 0, ~ 1}\mu(p_1^{b_1}p_2^{b_2}\cdots p_m^{b_m}) \\ &= \sum_{b_k = 0, ~ 1}\mu(p_1^{b_1})\mu(p_2^{b_2})\cdots\mu(p_m^{b_m}) \\ &= \sum_{b_k = 0, ~ 1}(-1)^{b_1 + b_2 + \cdots + b_m} \\ &= \sum_s^m\binom{m}{s}(-1)^s \\ &= (1 - 1)^m = 0 \end{aligned} (1∗μ)(n)=bk=0, 1∑μ(p1b1p2b2⋯pmbm)=bk=0, 1∑μ(p1b1)μ(p2b2)⋯μ(pmbm)=bk=0, 1∑(−1)b1+b2+⋯+bm=s∑m(sm)(−1)s=(1−1)m=0
其中最后一步是二项式定理。综合两种情况就得到要证的结果。
g = ( 1 ∗ f ) ⟺ f = ( μ ∗ g ) g = (1 * f) \iff f = (\mu * g) g=(1∗f)⟺f=(μ∗g)
两边同时对 μ \mu μ卷积,或者同时对 1 1 1卷积即可证明。
设数论函数 α ( n ) , β ( n ) \alpha(n), ~ \beta(n) α(n), β(n)满足 ( α ∗ β ) ( n ) = ϵ ( n ) (\alpha * \beta)(n) = \epsilon(n) (α∗β)(n)=ϵ(n), F , G F, ~ G F, G是定义在 [ 1 , + ∞ ) [1, +\infty) [1,+∞)上的函数,我们有
G ( x ) = ∑ 1 ≤ n ≤ x z ( n ) α ( n ) F ( x n ) ⟺ F ( x ) = ∑ 1 ≤ n ≤ x z ( n ) β ( n ) G ( x n ) G(x) = \sum_{1 \le n \le x}z(n)\alpha(n)F(\frac{x}{n}) \iff F(x) = \sum_{1 \le n \le x}z(n)\beta(n)G(\frac{x}{n}) G(x)=1≤n≤x∑z(n)α(n)F(nx)⟺F(x)=1≤n≤x∑z(n)β(n)G(nx)
其中 z ( n ) z(n) z(n)为一完全积性函数,即 z ( 1 ) = 1 z(1) = 1 z(1)=1且 z ( n m ) = z ( n ) z ( m ) z(nm) = z(n)z(m) z(nm)=z(n)z(m).
证明
∑ 1 ≤ n ≤ x z ( n ) β ( n ) G ( x n ) = ∑ 1 ≤ n ≤ x z ( n ) β ( n ) ∑ 1 ≤ m ≤ x n z ( m ) α ( m ) F ( x n m ) = ∑ 1 ≤ n ≤ x β ( n ) ∑ 1 ≤ m ≤ x n α ( m ) ∑ 1 ≤ r ≤ x [ r = m n ] z ( r ) F ( x r ) = ∑ 1 ≤ r ≤ x z ( r ) F ( x r ) ∑ 1 ≤ n ≤ x ∑ 1 ≤ m ≤ x n [ r = m n ] α ( m ) β ( n ) = ∑ 1 ≤ r ≤ x z ( r ) F ( x r ) ∑ m ∣ r α ( m ) β ( r m ) = ∑ 1 ≤ r ≤ x z ( r ) F ( x r ) ϵ ( r ) = F ( x ) \begin{aligned} \sum_{1 \le n \le x}z(n)\beta(n)G(\frac{x}{n}) &= \sum_{1 \le n \le x}z(n)\beta(n)\sum_{1 \le m \le \frac{x}{n}}z(m)\alpha(m)F(\frac{x}{nm}) \\ &= \sum_{1 \le n \le x}\beta(n)\sum_{1 \le m \le \frac{x}{n}}\alpha(m)\sum_{1\le r\le x}[r = mn]z(r)F(\frac{x}{r}) \\ &= \sum_{1\le r\le x}z(r)F(\frac{x}{r})\sum_{1 \le n \le x}\sum_{1 \le m \le \frac{x}{n}}[r = mn]\alpha(m)\beta(n) \\ &= \sum_{1\le r\le x}z(r)F(\frac{x}{r})\sum_{m \mid r}\alpha(m)\beta(\frac{r}{m}) \\ &= \sum_{1\le r\le x}z(r)F(\frac{x}{r})\epsilon(r) = F(x) \end{aligned} 1≤n≤x∑z(n)β(n)G(nx)=1≤n≤x∑z(n)β(n)1≤m≤nx∑z(m)α(m)F(nmx)=1≤n≤x∑β(n)1≤m≤nx∑α(m)1≤r≤x∑[r=mn]z(r)F(rx)=1≤r≤x∑z(r)F(rx)1≤n≤x∑1≤m≤nx∑[r=mn]α(m)β(n)=1≤r≤x∑z(r)F(rx)m∣r∑α(m)β(mr)=1≤r≤x∑z(r)F(rx)ϵ(r)=F(x)
令 α = 1 , β = μ \alpha = 1, ~ \beta = \mu α=1, β=μ,我们就有广义Möbius反演
G ( x ) = ∑ 1 ≤ n ≤ x z ( n ) F ( x n ) ⟺ F ( x ) = ∑ 1 ≤ n ≤ x z ( n ) μ ( n ) G ( x n ) G(x) = \sum_{1 \le n \le x}z(n)F(\frac{x}{n}) \iff F(x) = \sum_{1 \le n \le x}z(n)\mu(n)G(\frac{x}{n}) G(x)=1≤n≤x∑z(n)F(nx)⟺F(x)=1≤n≤x∑z(n)μ(n)G(nx)
类似地,我们有
G ( x ) = ∑ 1 ≤ n ≤ x z ( n ) F ( x n ) ⟺ F ( x ) = ∑ 1 ≤ n ≤ x z ( n ) μ ( n ) G ( x n ) G(x) = \sum_{1 \le n \le x}z(n)F(\sqrt[n]x) \iff F(x) = \sum_{1 \le n \le x}z(n)\mu(n)G(\sqrt[n]x) G(x)=1≤n≤x∑z(n)F(nx)⟺F(x)=1≤n≤x∑z(n)μ(n)G(nx)
证明是完全类似的。
黎曼也是从
ln ζ ( s ) = ∑ p ∑ n 1 n p n s \ln\zeta(s) = \sum_p\sum_n\frac{1}{np^{ns}} lnζ(s)=p∑n∑npns1
入手,可以证明 ln ζ ( s ) \ln\zeta(s) lnζ(s)在复平面上 ℜ ( s ) > 0 \Re(s) > 0 ℜ(s)>0的区域是绝对收敛的。定义
J ( x ) = π ( x ) + 1 2 π ( x ) + 1 3 π ( x 3 ) + ⋯ = ∑ n [ 1 n π ( x n ) ] J(x) = \pi(x) + \frac{1}{2}\pi(\sqrt x) + \frac{1}{3}\pi(\sqrt[3] x) + \cdots = \sum_n[\frac{1}{n}\pi(\sqrt[n]x)] J(x)=π(x)+21π(x)+31π(3x)+⋯=n∑[n1π(nx)]
称为黎曼素数计数函数。显然 J ( 0 ) = 0 J(0) = 0 J(0)=0,而后其每越过一个素数就增加 1 1 1,每越过一个素数的平方就增加 1 2 \frac{1}{2} 21,如此这般。在其不连续的点上,其值用 J ( x ) = 1 2 [ J ( x − ) + J ( x + ) ] J(x) = \frac{1}{2}[J(x^-) + J(x^+)] J(x)=21[J(x−)+J(x+)]定义。可以看出, J ( x ) J(x) J(x)和 π ( x ) \pi(x) π(x)之间的关系正是由上面提到的广义Möbius反演所联系
π ( x ) = ∑ n [ μ ( x ) n J ( x n ) ] \pi(x) = \sum_n[\frac{\mu(x)}{n}J(\sqrt[n]x)] π(x)=n∑[nμ(x)J(nx)]
借此函数,上式就可以表为积分形式,因为
∑ p ∑ n 1 n p n s = ∫ 0 ∞ ∑ n 1 n x n s ρ ( x ) d x = ∫ 0 ∞ t − s ∑ n 1 n d π ( t n ) = ∫ 0 ∞ t − s d J ( t ) \sum_p\sum_n\frac{1}{np^{ns}} = \int_0^\infty\sum_n\frac{1}{nx^{ns}}\rho(x)dx = \int_0^\infty t^{-s}\sum_n\frac{1}{n}d\pi(\sqrt[n]t) = \int_0^\infty t^{-s}dJ(t) p∑n∑npns1=∫0∞n∑nxns1ρ(x)dx=∫0∞t−sn∑n1dπ(nt)=∫0∞t−sdJ(t)
其中第二个等号处使用换元 x = t n x = \sqrt[n]t x=nt. 再进行一次分部积分便得
ln ζ ( s ) = s ∫ 0 ∞ J ( x ) x − s − 1 d x \ln\zeta(s) = s\int_0^\infty J(x)x^{-s - 1}dx lnζ(s)=s∫0∞J(x)x−s−1dx
Mellin变换的定义是
{ M f } ( s ) = φ ( s ) = ∫ 0 ∞ x s − 1 f ( x ) d x \{\mathcal Mf\}(s) = \varphi(s) = \int_0^\infty x^{s-1}f(x)dx {Mf}(s)=φ(s)=∫0∞xs−1f(x)dx
其逆变换为
{ M − 1 φ } ( x ) = f ( x ) = 1 2 π i ∫ c − i ∞ c + i ∞ x − s φ ( s ) d s \{\mathcal M^{-1}\varphi\}(x) = f(x) = \frac{1}{2\pi i}\int_{c - i\infty}^{c + i\infty}x^{-s}\varphi(s)ds {M−1φ}(x)=f(x)=2πi1∫c−i∞c+i∞x−sφ(s)ds
根据上述定义就有
ln ζ ( s ) s = { M J } ( − s ) \frac{\ln\zeta(s)}{s} = \{\mathcal MJ\}(-s) slnζ(s)={MJ}(−s)
于是根据逆变换得到
J ( x ) = 1 2 π i ∫ a − i ∞ a + i ∞ ln ζ ( z ) z x z d z J(x) = \frac{1}{2\pi i}\int_{a - i\infty}^{a + i\infty}\frac{\ln\zeta(z)}{z}x^zdz J(x)=2πi1∫a−i∞a+i∞zlnζ(z)xzdz
其中 z = − s , a = − c z = -s, ~ a = -c z=−s, a=−c为大于 1 1 1的实数。这就将 ζ \zeta ζ函数和素数分布规律 J J J明确地联系了起来。