因为X1,X2,…,Xn相互独立,所以
D(∑(i从1到n)aiXi) = ∑(i从1到n)D(aiXi) = ∑(i从1到n)ai^2 D(Xi) = ∑(i从1到n)ai^2 δi^2
设 L(a1,...,an,λ) = ∑(i从1到n)(aiδi)^2+λ(∑(i从1到n)ai-1),
当给定 a1,...,a(i-1),a(i+1),...,an,λ时,L是ai的二次函数,且开口向上.
于是在最小值处,有:
下面用 dL/dai 表示偏导数.
dL/dai = 2ai δi^2 + λ = 0 ,i = 1,...,n
==> -λ/2 = a1 δ1^2 = a1/(1/ δ1^2) = .= an/(1/ δn^2)
= (a1 + .+an)/((1/ δ1^2) + ...+(1/ δn^2))
= 1/ ((1/ δ1^2) + ...+(1/ δn^2))
==>
ai = -λ / (2δi^2) = 1/δi^2 * (-λ/2)= 1/δi^2 / ((1/ δ1^2) + ...+(1/ δn^2)) ,i = 1,2,...,n
当 ai ,i=1,...,n,为上值时,方差最小.