時間序列︰生成函數‧漸近展開︰當白努利遇上傅立葉《I》

如何求取白努利多項式周期函數 {\tilde {B}}_{n} (x) = B_n (x - \lfloor x \rfloor) 之傅立葉級數呢?

傅立葉級數

數學中,傅立葉級數Fourier series, /ˈfɔəri/)是把類似波的函數表示成簡單正弦波的方式。更正式地說,它能將任何周期函數或周期訊號分解成一個(可能由無窮個元素組成的)簡單振盪函數的集合,即正弦函數餘弦函數(或者,等價地使用複指數)。離散時間傅立葉轉換是一個周期函數,通常用定義傅立葉級數的項進行定義。另一個應用的例子是Z轉換,將傅立葉級數簡化為特殊情形 |z|=1。傅立葉級數也是取樣定理原始證明的核心。傅立葉級數的研究是傅立葉分析的一個分支。

歷史

傅立葉級數得名於法國數學家約瑟夫·傅立葉(1768年–1830年),他提出任何函數都可以展開為三角級數。此前數學家如拉格朗日等已經找到了一些非周期函數的三角級數展開,而認定一個函數有三角級數展開之後,通過積分方法計算其係數的公式,歐拉達朗貝爾克萊羅早已發現,傅立葉的工作得到了丹尼爾·伯努利的贊助[1]。傅立葉介入三角級數用來解熱傳導方程,其最初論文在1807年經拉格朗日拉普拉斯勒讓德評審後被拒絕出版,他的現在被稱為傅里葉逆轉定理的理論後來發表於1820年的《熱的解析理論》中。將周期函數分解為簡單振盪函數的總和的最早想法,可以追溯至公元前3世紀古代天文學家的均輪和本輪學說。

傅立葉級數在數論組合數學訊號處理、機率論、統計學、密碼學、聲學、光學等領域都有著廣泛的應用。

定義

在這一節中,s(x) 表示實變量 x 的一個函數,且 s 在 [x0x0 + P] 上可積,x0 和 P 為實數。我們將嘗試用諧波關係的正弦函數的無窮和或級數來表示該區間內的  s 。在區間外,級數以 P 為周期(頻率為 1/P)。若 s 也具有該性質,則它的近似在整個實數線上有效。我們可以從有限求和(或部分和)開始:

  </span

  s_{N}(x)  為周期為 P 的周期函數。運用恆等式:

  </span
  </span

函數 s(x) (紅色)是六個不同幅度的諧波關係的正弦函數的和。它們的和叫做傅立葉級數。傅立葉轉換 S(f) (藍色),針對幅度與頻率進行描繪,顯示出6種頻率和它們的幅度。

我們還可以用這些等價形式書寫這個函數:

{\begin{aligned}s_{N}(x)&={\frac {a_{0}}{2}}+\sum _{{n=1}}^{N}\left(\overbrace {a_{n}}^{{A_{n}\sin(\phi _{n})}}\cos({\tfrac {2\pi nx}{P}})+\overbrace {b_{n}}^{{A_{n}\cos(\phi _{n})}}\sin({\tfrac {2\pi nx}{P}})\right)\\&=\sum _{{n=-N}}^{N}c_{n}\cdot e^{{i{\tfrac {2\pi nx}{P}}}},\end{aligned}}

其中:

  c_{n}\ {\stackrel {{\mathrm {def}}}{=}}\ {\begin{cases}{\frac {A_{n}}{2i}}e^{{i\phi _{n}}}={\frac {1}{2}}(a_{n}-ib_{n})&{\text{for }}n>0\\{\frac {1}{2}}a_{0}&{\text{for }}n=0\\c_{{|n|}}^{*}&{\text{for }}n<0.\end{cases}}

當係數(即傅立葉係數)以下面方式計算時:[2]

  a_{n}={\frac {2}{P}}\int _{{x_{0}}}^{{x_{0}+P}}s(x)\cdot \cos({\tfrac {2\pi nx}{P}})\ dx
  b_{n}={\frac {2}{P}}\int _{{x_{0}}}^{{x_{0}+P}}s(x)\cdot \sin({\tfrac {2\pi nx}{P}})\ dx
 c_{n}={\frac {1}{P}}\int _{{x_{0}}}^{{x_{0}+P}}s(x)\cdot e^{{-i{\tfrac {2\pi nx}{P}}}}\ dx,

  s_{N}(x)  在   [x_{0},\ x_{0}+P] 近似了  s(x) ,該近似程度會隨著 N → ∞ 逐漸改善。這個無窮和  s_{{\infty }}(x) 叫做  s 的傅立葉級數表示。在工程應用中,一般假定傅立葉級數除了在不連續點以外處處收斂,原因是工程上遇到的函數比數學家提供的這個假定的反例表現更加良好。特別地,傅立葉級數絕對收斂且均勻收斂於 s(x),只要在 s(x) 的導數(或許不會處處存在)是平方可積的。[3]  如果一個函數在區間 [x0, x0+P]上是平方可積的,那麼此傅立葉級數在幾乎所有點都收斂於該函數。傅立葉級數的收斂性取決於函數有限數量的極大值和極小值,這就是通常稱為傅立葉級數的狄利克雷條件。參見傅立葉級數的收斂性之一。對於廣義函數或分布也可以用範數或弱收斂定義傅立葉係數.

 

且讓我們介紹數學分析裡常用之技巧『分部積分法』︰

Integration by parts

In calculus, and more generally in mathematical analysis, integration by parts or partial integration is a theorem that relates the integral of a product of functions to the integral of their derivative and antiderivative. It is frequently used to transform the antiderivative of a product of functions into an antiderivative for which a solution can be more easily found. The rule can be derived in one line simply by integrating the product rule of differentiation.

If u = u(x) and du = u′(xdx, while v = v(x) and dv = v′(xdx, then integration by parts states that:

{\displaystyle {\begin{aligned}\int _{a}^{b}u(x)v'(x)\,dx&=[u(x)v(x)]_{a}^{b}-\int _{a}^{b}u'(x)v(x)dx\\&=u(b)v(b)-u(a)v(a)-\int _{a}^{b}u'(x)v(x)\,dx\end{aligned}}}

or more compactly:

\int u\,dv=uv-\int v\,du.\!

More general formulations of integration by parts exist for the Riemann–Stieltjes and Lebesgue–Stieltjes integrals. The discrete analogue for sequences is called summation by parts.

Visualization

Graphical interpretation of the theorem. The pictured curve is parametrized by the variable t.

Define a parametric curve by (x, y) = (f(t), g(t)). Assuming that the curve is locally one-to-one, we can define

  x(y)=f(g^{-1}(y))
  y(x)=g(f^{-1}(x))

The area of the blue region is

  A_{1}=\int _{y_{1}}^{y_{2}}x(y)dy

Similarly, the area of the red region is

  A_{2}=\int _{x_{1}}^{x_{2}}y(x)dx

The total area A1 + A2 is equal to the area of the bigger rectangle, x2y2, minus the area of the smaller one, x1y1:

  \overbrace {\int _{y_{1}}^{y_{2}}x(y)dy} ^{A_{1}}+\overbrace {\int _{x_{1}}^{x_{2}}y(x)dx} ^{A_{2}}={\biggl .}x.y(x){\biggl |}_{x1}^{x2}={\biggl .}y.x(y){\biggl |}_{y1}^{y2}

Assuming the curve is smooth within a neighborhood, this generalizes to indefinite integrals:

  \int xdy+\int ydx=xy

Rearranging:

  \int xdy=xy-\int ydx

Thus integration by parts may be thought of as deriving the area of the blue region from the total area and that of the red region.

This visualisation also explains why integration by parts may help find the integral of an inverse function f−1(x) when the integral of the function f(xv) is known. Indeed, the functions x(y) and y(x) are inverses, and the integral ∫x dy may be calculated as above from knowing the integral ∫y dx.

 

並以一個例子展現它的威力︰

\int_{0}^{1} B_n (x) B_m(x) dx

= \frac{1}{n+1} B_{n+1} (x) B_m (x) |_{0}^{1} - \frac{m}{n+1} \int_{0}^{1} B_{n+1} (x) B_{m-1} (x) dx

= - \frac{m}{n+1} \int_{0}^{1} B_{n+1} (x) B_{m-1} (x) dx

= \cdots

= {(-1)}^{m-1} \frac{m \cdot (m-1) \cdots \cdot 2}{(n+1) \cdot (n+2) \cdots \cdot (n+m)} B_{n+m} (x) B_1 (x) |_{0}^{1} - {(-1)}^{m-1}  \frac{m \cdot (m-1) \cdots \cdot 2 \cdot 1}{(n+1) \cdot (n+2) \cdots \cdot (n+m)} \int_{0}^{1} B_{n+m} (x) B_0 (x) dx

= {(-1)}^{m-1} \frac{1}{\binom{n+m}{n}} \left( B_{n+m} (1) \cdot \frac{1}{2} - B_{n+m} (0) \cdot (-\frac{1}{2}) \right) - 0

= \frac{{(-1)}^{m-1}}{\binom{n+m}{n}} b_{n+m}

\int_{0}^{1} {\left( \frac{B_n (x)}{n!} \right) }^2 dx = \frac{| b_{2n} | }{(2n)!}

 

回顧白努利多項式的

積分表達式

\int_{0}^{1} B_n (x) dx = 0, \ n \ge 1 ,以及

端點關係

B_n (1) = B_n (0), \ n \neq 1

建議我們將 {\tilde{B}}_1 (x) 分開考慮。茲計算如下︰

c_0 ({\tilde{B}}_1 (x) ) = \int_{0}^{1} B_1 (x) dx = 0

c_k ({\tilde{B}}_1 (x) ) = \int_{0}^{1} B_1 (x) e^{-i 2 \pi k \cdot x }dx

= \int_{0}^{1} (x - \frac{1}{2}) e^{-i 2 \pi k \cdot x }dx

= - \frac{ e^{-i 2 \pi k \cdot x }}{i 2 \pi k} (x - \frac{1}{2}) |_{0}^{1} + \frac{1}{i 2 \pi k} \int_{0}^{1} e^{-i 2 \pi k \cdot x }dx

= \left[ -\frac{ e^{-i 2 \pi k \cdot 1 }}{i 2 \pi k} (1 - \frac{1}{2}) + \frac{ e^{-i 2 \pi k \cdot 0 }}{i 2 \pi k} (0 - \frac{1}{2})\right] + 0

= - \frac{1}{i 2 \pi k} 。所以

\therefore {\tilde{B}}_1 (x) = Re \left[ \sum \limits_{k \in Z, k \neq 0}^{\infty}  - \frac{1}{i 2 \pi k}e^{+i 2 \pi k \cdot x} \right]

= Re \left[ \sum \limits_{k \in Z, k \neq 0}^{\infty}  - \frac{1}{i 2 \pi k} (  \cos(2 \pi k cdot x) + i \sin(2 \pi k \cdot x)) \right]

= - \frac{1}{\pi} \sum \limits_{k=1}^{\infty}  \frac{\sin(2 \pi k \cdot x)}{k}

 

請讀者特別注意上式左邊

\because x \in [0, 1), {\tilde{B}}_1 (x) = B_1 (x) = x - \frac{1}{2}, \ \therefore  {\tilde{B}}_1 (0) = - \frac{1}{2} ,右邊卻因為 \sin(2 \pi k \cdot 0 ) = 0 ,故為零。實由 B_1 (1) = \frac{1}{2} \neq - \frac{1}{2} = B_1 (0) 所生,或說不『連續性』引發之現象也! 故而有『Analytic continuation0 = \frac{(\frac{1}{2}) + (- \frac{1}{2})}{2} 之論哩︰

解析延拓

解析延拓數學上將解析函數從較小定義域拓展到更大定義域的方法。透過此方法,一些原先發散級數在新的定義域可具有迥異而有限的值。其中最知名的例子為Γ函數黎曼ζ函數

初步闡述

自然對數虛部之解析延拓

f為一解析函數,定義於複平面C中之一開子集 U,而VC中一更大且包含U之開子集。F為定義於V之解析函數,並使

  \displaystyle F(z)=f(z)\qquad \forall z\in U,

F稱為f之解析延拓。換過來說,將F函數限制在U則得到原先的f函數。

解析延拓具有唯一性:

V為兩解析函數F1F2連通定義域,並使V包含U;若在U中所有的z使得

F1(z) = F2(z) = f(z),

則在V中所有點

F1 = F2

此乃因 F1 − F2亦為一解析函數,其值於f的開放連通定義域U上為0,必導致整個定義域上的值皆為0。此為全純函數惟一性定理的直接結果。

 

同理當 n \ge 2 時︰

c_0 ({\tilde{B}}_n (x) ) = \int_{0}^{1} B_n (x) dx = 0

c_k ({\tilde{B}}_n (x) ) = \int_{0}^{1} B_n (x) e^{-i 2 \pi k \cdot x }dx

= - \frac{ e^{-i 2 \pi k \cdot x }}{i 2 \pi k} B_n (x) |_{0}^{1} + \frac{1}{i 2 \pi k} \int_{0}^{1} B_{n}^{'} (x) e^{-i 2 \pi k \cdot x }dx

= 0 + \frac{n}{i 2 \pi k} \int_{0}^{1} B_{n-1} (x) e^{-i 2 \pi k \cdot x } dx

= \frac{n}{i 2 \pi k}  c_k ({\tilde{B}}_{n-1} (x) ) 。故可用數學歸納法得

c_k ({\tilde{B}}_n (x) ) = - \frac{n!}{{(i 2 \pi k )}^n}, \ k \neq 0 。因此

B_{2n} (x) = {(-1)}^{n+1} \frac{2 (2n)!}{{(2 \pi )}^{2n}} \sum \limits_{k=1}^{\infty}  \frac{\cos(2 \pi k \cdot x)}{k^{2n}}

B_{2n+1} (x) = {(-1)}^{n+1} \frac{2 (2n+1)!}{{(2 \pi )}^{2n+1}} \sum \limits_{k=1}^{\infty}  \frac{\sin(2 \pi k \cdot x)}{k^{2n+1}}

 

恰可以三角函數性質歸結為☆

Fourier series

The Fourier series of the Bernoulli polynomials is also a Dirichlet series, given by the expansion

B_n(x) = -\frac{n!}{(2\pi i)^n}\sum_{k\not=0 }\frac{e^{2\pi ikx}}{k^n}= -2 n! \sum_{k=1}^{\infty} \frac{\cos\left(2 k \pi x- \frac{n \pi} 2 \right)}{(2 k \pi)^n}.

Note the simple large n limit to suitably scaled trigonometric functions.