Let X be a random variable with distribution g(x)=ΣNi=1λifi(x), where each fi is a probability distribution for the RV Xi (with mean μi and variance σ2i), λi≥0 and ΣNi=1λi=1.
We determine the mean, E[X]=μ and variance, V[X]=σ2, of X in terms of those of the Xi.
We have
μ=E[X]=E[ΣNi=1λiXi]=ΣλiE[Xi]=ΣNi=1λiμi
by linearity.
Then on expanding the square we have
σ2=V[X]=E[(X−μ)2]=E[X2]−μ2,
but that isn't really any help to us, because we haven't determined E[X2].
Alternatively,
σ2=∫dx(x−μ)2f(x)=∫dx(x−μ)2Σiλifi(x)=Σiλi∫(x−μ)2fi(x)
=Σiλi∫dxfi(x)(x−μi+μi−μ)2
=Σiλi∫dxfi(x)[(x−μi)2+2(x−μi)(μi−μ)+(μi−μ)2]
=Σiλi[∫dxfi(x)(x−μi)2+2(μi−μ)∫dxfi(x)(x−μi)+(μi−μ)2∫dxfi(x)]
=ΣNi=1λi(E[(Xi−μi)2]+2(μi−μ)(E[Xi]−μi)+(μi−μ)2)
=ΣNi=1λi(V[Xi]+(μi−μ)2)
≥ΣiλiV[Xi]
No comments:
Post a Comment