題目
Problem
3. Let S ( a , b ) = ∑ i ( y i − a − b x i ) 2 S(a, b) = \sum_i (y_i - a - b x_i)^2 S ( a , b ) = ∑ i ( y i − a − b x i ) 2 , where ∑ i x i 2 = n , ∑ i x i = n / 2 \sum_i x_i^2 = n, \ \sum_i x_i = n/2 ∑ i x i 2 = n , ∑ i x i = n /2 .
(1) Let ( a 1 , b 1 ) (a_1, b_1) ( a 1 , b 1 ) be the minimizer of S ( a , b ) S(a, b) S ( a , b ) . Find ( a 1 , b 1 ) (a_1, b_1) ( a 1 , b 1 ) . (6%)
(2) Under the constraint a + b = c a + b = c a + b = c , find the minimizer of S ( a , b ) S(a, b) S ( a , b ) in terms of ( a 1 , b 1 ) (a_1, b_1) ( a 1 , b 1 ) and c c c . (6%)
Ps., y i y_i y i is a known number for i = 1 , 2 , … , n i = 1, 2, \dots, n i = 1 , 2 , … , n .
解答
(1) 求解無約束最小化最優解 ( a 1 , b 1 ) (a_1, b_1) ( a 1 , b 1 )
我們要對偏偏導數設為零:
∂ S ∂ a = − 2 ∑ i = 1 n ( y i − a − b x i ) = 0 ⟹ ∑ i = 1 n y i = n a + b ∑ i = 1 n x i — (1) \frac{\partial S}{\partial a} = -2\sum_{i=1}^n (y_i - a - b x_i) = 0 \implies \sum_{i=1}^n y_i = na + b\sum_{i=1}^n x_i \quad \text{--- (1)} ∂ a ∂ S = − 2 i = 1 ∑ n ( y i − a − b x i ) = 0 ⟹ i = 1 ∑ n y i = na + b i = 1 ∑ n x i — (1)
∂ S ∂ b = − 2 ∑ i = 1 n x i ( y i − a − b x i ) = 0 ⟹ ∑ i = 1 n x i y i = a ∑ i = 1 n x i + b ∑ i = 1 n x i 2 — (2) \frac{\partial S}{\partial b} = -2\sum_{i=1}^n x_i(y_i - a - b x_i) = 0 \implies \sum_{i=1}^n x_i y_i = a\sum_{i=1}^n x_i + b\sum_{i=1}^n x_i^2 \quad \text{--- (2)} ∂ b ∂ S = − 2 i = 1 ∑ n x i ( y i − a − b x i ) = 0 ⟹ i = 1 ∑ n x i y i = a i = 1 ∑ n x i + b i = 1 ∑ n x i 2 — (2)
將已知條件 ∑ i = 1 n x i = n 2 \sum_{i=1}^n x_i = \frac{n}{2} ∑ i = 1 n x i = 2 n 與 ∑ i = 1 n x i 2 = n \sum_{i=1}^n x_i^2 = n ∑ i = 1 n x i 2 = n 代回方程組:
{ n a + n 2 b = ∑ y i n 2 a + n b = ∑ x i y i \begin{cases}
n a + \frac{n}{2} b = \sum y_i \\[2mm]
\frac{n}{2} a + n b = \sum x_i y_i
\end{cases} ⎩ ⎨ ⎧ na + 2 n b = ∑ y i 2 n a + nb = ∑ x i y i
我們將兩式同除以 n n n 簡化:
{ a + 1 2 b = y ˉ ( 其中 y ˉ = 1 n ∑ y i ) 1 2 a + b = 1 n ∑ x i y i \begin{cases}
a + \frac{1}{2} b = \bar{y} \quad (\text{其中 } \bar{y} = \frac{1}{n}\sum y_i) \\[2mm]
\frac{1}{2} a + b = \frac{1}{n}\sum x_i y_i
\end{cases} ⎩ ⎨ ⎧ a + 2 1 b = y ˉ ( 其中 y ˉ = n 1 ∑ y i ) 2 1 a + b = n 1 ∑ x i y i
我們使用消元法解此線性方程組:
第一式乘以 2 減去第二式:
2 a + b − ( 1 2 a + b ) = 2 y ˉ − 1 n ∑ x i y i ⟹ 3 2 a = 2 y ˉ − 1 n ∑ x i y i 2a + b - \left(\frac{1}{2}a + b\right) = 2\bar{y} - \frac{1}{n}\sum x_i y_i \implies \frac{3}{2}a = 2\bar{y} - \frac{1}{n}\sum x_i y_i 2 a + b − ( 2 1 a + b ) = 2 y ˉ − n 1 ∑ x i y i ⟹ 2 3 a = 2 y ˉ − n 1 ∑ x i y i
a 1 = 4 y ˉ 3 − 2 3 n ∑ x i y i = 4 ∑ y i − 2 ∑ x i y i 3 n a_1 = \frac{4\bar{y}}{3} - \frac{2}{3n}\sum x_i y_i = \frac{4\sum y_i - 2\sum x_i y_i}{3n} a 1 = 3 4 y ˉ − 3 n 2 ∑ x i y i = 3 n 4 ∑ y i − 2 ∑ x i y i
第二式乘以 2 減去第一式:
a + 2 b − ( a + 1 2 b ) = 2 n ∑ x i y i − y ˉ ⟹ 3 2 b = 2 n ∑ x i y i − y ˉ a + 2b - \left(a + \frac{1}{2}b\right) = \frac{2}{n}\sum x_i y_i - \bar{y} \implies \frac{3}{2}b = \frac{2}{n}\sum x_i y_i - \bar{y} a + 2 b − ( a + 2 1 b ) = n 2 ∑ x i y i − y ˉ ⟹ 2 3 b = n 2 ∑ x i y i − y ˉ
b 1 = 4 ∑ x i y i − 2 ∑ y i 3 n b_1 = \frac{4\sum x_i y_i - 2\sum y_i}{3n} b 1 = 3 n 4 ∑ x i y i − 2 ∑ y i
(2) 求解在約束 a + b = c a + b = c a + b = c 下的最優解
我們可以使用拉格朗日乘子法 (Lagrange Multipliers) 來求解此有約束的最優化問題。
定義拉格朗日函數:
L ( a , b , μ ) = ∑ i = 1 n ( y i − a − b x i ) 2 + μ ( a + b − c ) L(a, b, \mu) = \sum_{i=1}^n (y_i - a - b x_i)^2 + \mu (a + b - c) L ( a , b , μ ) = i = 1 ∑ n ( y i − a − b x i ) 2 + μ ( a + b − c )
對 a a a 與 b b b 求偏導數並令其為零:
∂ L ∂ a = − 2 ∑ i = 1 n ( y i − a − b x i ) + μ = 0 ⟹ 2 ( ∑ y i − n a − b ∑ x i ) = μ — (3) \frac{\partial L}{\partial a} = -2\sum_{i=1}^n (y_i - a - b x_i) + \mu = 0 \implies 2\left( \sum y_i - na - b\sum x_i \right) = \mu \quad \text{--- (3)} ∂ a ∂ L = − 2 i = 1 ∑ n ( y i − a − b x i ) + μ = 0 ⟹ 2 ( ∑ y i − na − b ∑ x i ) = μ — (3)
∂ L ∂ b = − 2 ∑ i = 1 n x i ( y i − a − b x i ) + μ = 0 ⟹ 2 ( ∑ x i y i − a ∑ x i − b ∑ x i 2 ) = μ — (4) \frac{\partial L}{\partial b} = -2\sum_{i=1}^n x_i(y_i - a - b x_i) + \mu = 0 \implies 2\left( \sum x_i y_i - a\sum x_i - b\sum x_i^2 \right) = \mu \quad \text{--- (4)} ∂ b ∂ L = − 2 i = 1 ∑ n x i ( y i − a − b x i ) + μ = 0 ⟹ 2 ( ∑ x i y i − a ∑ x i − b ∑ x i 2 ) = μ — (4)
將式 (3) 與 (4) 聯立消去乘子 μ \mu μ :
∑ y i − n a − b ∑ x i = ∑ x i y i − a ∑ x i − b ∑ x i 2 \sum y_i - na - b\sum x_i = \sum x_i y_i - a\sum x_i - b\sum x_i^2 ∑ y i − na − b ∑ x i = ∑ x i y i − a ∑ x i − b ∑ x i 2
我們代入已知條件 ∑ x i = n 2 \sum x_i = \frac{n}{2} ∑ x i = 2 n 與 ∑ x i 2 = n \sum x_i^2 = n ∑ x i 2 = n :
∑ y i − n a − n 2 b = ∑ x i y i − n 2 a − n b \sum y_i - na - \frac{n}{2}b = \sum x_i y_i - \frac{n}{2}a - nb ∑ y i − na − 2 n b = ∑ x i y i − 2 n a − nb
∑ y i − ∑ x i y i = n 2 a − n 2 b = n 2 ( a − b ) \sum y_i - \sum x_i y_i = \frac{n}{2}a - \frac{n}{2}b = \frac{n}{2}(a - b) ∑ y i − ∑ x i y i = 2 n a − 2 n b = 2 n ( a − b )
我們兩側同除以 n n n :
∑ y i n − ∑ x i y i n = 1 2 ( a − b ) — (5) \frac{\sum y_i}{n} - \frac{\sum x_i y_i}{n} = \frac{1}{2}(a - b) \quad \text{--- (5)} n ∑ y i − n ∑ x i y i = 2 1 ( a − b ) — (5)
注意到無約束最優解 ( a 1 , b 1 ) (a_1, b_1) ( a 1 , b 1 ) 滿足:
∑ y i n = a 1 + 1 2 b 1 \displaystyle \frac{\sum y_i}{n} = a_1 + \frac{1}{2}b_1 n ∑ y i = a 1 + 2 1 b 1
∑ x i y i n = 1 2 a 1 + b 1 \displaystyle \frac{\sum x_i y_i}{n} = \frac{1}{2}a_1 + b_1 n ∑ x i y i = 2 1 a 1 + b 1
我們將這兩個關係代入式 (5) 的左側:
( a 1 + 1 2 b 1 ) − ( 1 2 a 1 + b 1 ) = 1 2 ( a − b ) \left( a_1 + \frac{1}{2}b_1 \right) - \left( \frac{1}{2}a_1 + b_1 \right) = \frac{1}{2}(a - b) ( a 1 + 2 1 b 1 ) − ( 2 1 a 1 + b 1 ) = 2 1 ( a − b )
1 2 a 1 − 1 2 b 1 = 1 2 ( a − b ) ⟹ a − b = a 1 − b 1 — (6) \frac{1}{2}a_1 - \frac{1}{2}b_1 = \frac{1}{2}(a - b) \implies a - b = a_1 - b_1 \quad \text{--- (6)} 2 1 a 1 − 2 1 b 1 = 2 1 ( a − b ) ⟹ a − b = a 1 − b 1 — (6)
現在,我們聯立式 (6) 與約束條件 a + b = c a + b = c a + b = c :
{ a − b = a 1 − b 1 a + b = c \begin{cases}
a - b = a_1 - b_1 \\
a + b = c
\end{cases} { a − b = a 1 − b 1 a + b = c
相加可解得 a a a :
2 a = a 1 − b 1 + c ⟹ a = a 1 − b 1 + c 2 2a = a_1 - b_1 + c \implies a = \frac{a_1 - b_1 + c}{2} 2 a = a 1 − b 1 + c ⟹ a = 2 a 1 − b 1 + c
相減可解得 b b b :
2 b = c − ( a 1 − b 1 ) ⟹ b = − a 1 + b 1 + c 2 2b = c - (a_1 - b_1) \implies b = \frac{-a_1 + b_1 + c}{2} 2 b = c − ( a 1 − b 1 ) ⟹ b = 2 − a 1 + b 1 + c
結論:
無約束最優解為 a 1 = 4 ∑ y i − 2 ∑ x i y i 3 n \displaystyle a_1 = \frac{4\sum y_i - 2\sum x_i y_i}{3n} a 1 = 3 n 4 ∑ y i − 2 ∑ x i y i , b 1 = 4 ∑ x i y i − 2 ∑ y i 3 n \displaystyle b_1 = \frac{4\sum x_i y_i - 2\sum y_i}{3n} b 1 = 3 n 4 ∑ x i y i − 2 ∑ y i 。
有約束最優解為 a = a 1 − b 1 + c 2 \displaystyle a = \frac{a_1 - b_1 + c}{2} a = 2 a 1 − b 1 + c , b = − a 1 + b 1 + c 2 \displaystyle b = \frac{-a_1 + b_1 + c}{2} b = 2 − a 1 + b 1 + c 。