「統計解析のための線形代数」復習筆記1

和記號\(\sum\)

  • \(\sum\) 的性質 (1) 下標(添字) \(x_1 + x_2 + x_3 + \dots + x_n\) 記作如下:\[\sum_{i=1}^{n}x_i\]
    • \(\sum_{i=1}^{n}x_i\) 中的\(i\) 稱爲dummy index
    • 可以簡略寫爲:\(\sum x\) 或者 \(\sum_1 x_i\), \(\sum x_i\)
  • \(\sum\) 的性質 (2)
    • \[\begin{equation} \sum_{i=1}^{n}(ax_i + by_i)= a\sum_{i=1}^{n}x_i + b\sum_{i=1}^{n}y_i \tag{1} \end{equation}\]
    • \(\sum_{i=1}^{n}ax_i = a\sum_{i=1}^{n}x_i\) 常數(定数)可以提前
    • \(\sum_{i=1}^{n}a = na\)
    • \(\sum_{i=1}^{n}1 = n\)
    • \(\sum_{i=1}^{n}(ax_i+b) = a\sum_{i=1}^{n}x_i + nb\)
    • 公式(1)的應用: \[ \begin{aligned} \sum_{i=1}^{n}(ax_i -by_i)^2 &= \sum_{i=1}^{n}(a^2x_i^2 - 2abx_iy_i + b^2y_i^2) \\ &= \sum_{i=1}^{n}a^2x_i^2 -\sum_{i=1}^{n}2abx_iy_i + \sum_{i=1}^{n}b^2y_i^2 \\ &= a^2\sum_{i=1}^{n}x_i^2 - 2ab\sum_{i=1}^{n}x_iy_i + b^2\sum_{i=1}^{n}y_i^2 \end{aligned} \]
    • 但是,乘法或平方有如下性質,計算方差(分散)或者相關系數時需要注意:\[\sum_{i=1}^{n}x_i^2 \neq (\sum_{i=1}^{n}x_i)^2\] 以及 \[\sum_{i=1}^{n}x_iy_i \neq (\sum_{i=1}^{n}x_i)(\sum_{i=1}^{n}y_i)\]
    • 自然數的冪運算之和(冪[べき]乗の和)的公式: \[ \begin{aligned} 1+2+3+\dots+n &= \sum_{t=1}^{n}t = \frac{n(n+1)}{2}\\ 1^2+2^2+3^2+\dots+n^2 &= \sum_{t=1}^{n}t^2 = \frac{n(n+1)(2n+1)}{6} \\ 1^3+2^3+3^3+\dots+n^3 &= \sum_{t=1}^{n}t^3 = {\frac{n(n+1)}{2}}^2 \\ 1^4+2^4+3^4+\dots+n^4 &= \sum_{t=1}^{n}t^4 = \frac{n(n+1)(2n+1)(3n^2+3n-1)}{30} \end{aligned} \]
      上面的公式將會應用在時間序列分析,斯皮尔曼等级相关系数(スピアマンの順位相関係数)的定義公式的推導。

\(\sum\)式子變形成普通計算式:

  1. \[\sum_{i=1}^{n}f_{ij} = f_{i1} + f_{i2} + f_{i3} + \dots + f_{in}\] 此式子也常寫作\(f_{i+}\), 或者\(f_{i\cdot}\)
  2. \[\sum_{i=1}^{m}f_{ij} = f_{1i} + f_{2i} + f_{3i} + \dots + f_{mj}\] 此式子也常寫作\(f_{+j}\), 或者\(f_{\cdot j}\)
  3. \[\sum_{i=1}^{2}\sum_{j=1}^{3}x_{ij} = \sum_{i=1}^{2}(\sum_{i=j}^{3}x_{ij}) = \sum_{i=1}^{2}(x_{i1} + x_{i2} + x_{i3}) = (x_{11} + x_{12} + x_{13}) + (x_{21} + x_{22} + x_{23})\] 此式子也可以寫作\(x_{++}\), 或者\(x_{\cdot\cdot}\)。另外,中間的式子如果是\(\sum_{j=1}^{3}(\sum_{i=1}^{2}x_{ij})\)也可以成立,過程如下:\[\sum_{j=1}^{3}(\sum_{i=1}^{2}x_{ij})=\sum_{j=1}^{3}(x_{1j} + x_{2j}) = (x_{11} + x_{21}) + (x_{12} + x_{22}) + (x_{13} + x_{23})\]

  1. \[ \begin{aligned} \sum_{i=1}^2\sum_{j=1}^2a_{ij}x_ix_j &= \sum_{i=1}^2(\sum_{j=1}^2a_{ij}x_ix_j) \\ &= \sum_{i=1}^2({\sum_{j=1}^2a_{ij}x_j)x_i} \\ &= \sum_{i=1}^2(a_{i1}x_1 + a_{i2}x_2)x_i \\ &= (a_{11}x_1 + a_{12}x_2)x_1 + (a_{21}x_1 + a_{22}x_2)x_2 \\ &= a_{11}x_1^2 + (a_{12} + a_{21})x_1x_2 + a_{22}x_2^2 \end{aligned} \]

  1. \[ \begin{aligned} \sum_{k=1}^3\left\{(\sum_{i=1}^2b_ix_{ik})(\sum_{j=1}^2b_jx_{jk})\right\} &= \sum_{k=1}^3\left\{(b_1x_{1k} + b_2x_{2k})(b_1x_{1k} + b_2x_{2k})\right\} \\ &= \sum_{k=1}^3(b_1x_{1k} + b_2x_{2k})^2 \\ &= (b_1x_{11} + b_2x_{21})^2 + (b_1x_{12} + b_2x_{22})^2 + (b_1x_{13} + b_2x_{23})^2 \end{aligned} \]

  1. \(\mathop{\sum\limits^3\sum\limits^3}\limits_{i<j}e_{ij}\) 會變成怎樣的式子呢? 滿足 \(i<j (i = 1,2,3; j = 1,2,3)\) 條件的 \(i,j\), 有且僅有 \((1,2),(1,3),(2,3)\),故 \(\mathop{\sum\limits^3\sum\limits^3}\limits_{i<j}e_{ij} = e_{12} + e_{13} + e_{23}\)

  1. 那麼\(\mathop{\sum\limits^3\sum\limits^3}\limits_{i\neq j}e_{ij}\)又會變成怎樣的式子呢? 滿足 \(i\neq j (i = 1,2,3; j = 1,2,3)\)\((i,j)\) 有6種組合:\((1,2),(2,1),(1,3),(3,1),(2,3),(3,2)\), 故\(\mathop{\sum\limits^3\sum\limits^3}\limits_{i\neq j}e_{ij} = e_{12} + e_{21} + e_{13} + e_{31} + e_{23} + e_{32}\)

加法算式變形爲\(\sum\)

在多元變量分析(多変量解析)中,與前項相比,加法算式變形成爲\(\sum\)式子更加重要。也就是說,以前項計算爲例的話,作爲答案的計算式如果放在題幹,反向求解\(\sum\)式的過程更加常用。簡單練習一下吧:

  1. 將計算式\(a_1x_1^2 + a_2x_2^2 + a_3x_3^2 + a_4x_4^2 + a_5x_5^2\)改寫成\(\sum\)式:
    • 先寫下:\(\sum\)
    • 再寫各個單元的共通部分\((a,x^2)\)\(\sum ax^2\)
    • 各單元不同的僅爲下標: \(\sum a_ix_i^2\)
    • 注意到下標的變化規律爲\(1\)\(5\)之間的整數,故在\(\sum\)符號的上部寫上\(5\),下部寫上\(i=1\): \(\sum\limits_{i=1}^5a_ix_i^2\) (答)
  2. 將計算式\(f_2(x_2 - \bar{x})^2 + f_3(x_3 - \bar{x})^2 + f_4(x_4 - \bar{x})^2 + f_5(x_5 - \bar{x})^2\)改寫成\(\sum\)式:
    • \(\sum\)
    • \(\sum f(x-\bar{x})^2\)
    • \(\sum f_i(x_i - \bar{x})^2\)
    • \(\sum\limits_{i=2}^5f_i(x_i - \bar{x})^2\) (答)

提問:

我們現在了解了可以使用簡單的\(\sum\)符號來表達復雜有規律的加法算式。請問有沒有相應的記號來代表乘法?

當然有。用希臘字母\(\pi\)的大寫\(\Pi\)來表示。例如: \[x_1x_2x_3x_4x_5 = \prod_{i=1}^5x_i\] \[x_1x_2x_3\dots x_n = \prod_{i=1}^nx_i\]

乘法記號可以證明下面的等式成立: \[\prod_{i=1}^nx_i^2 = (\prod_{i=1}^nx_i)^2\] \[\log(\prod_{i=1}^nx_i) = \prod_{i=1}^n\log x_i\]

Avatar
王 超辰 - Chaochen Wang
Real World Evidence Scientist

All models are wrong, but some are useful.

comments powered by Disqus