Prove that $\mbox{Var}_g[\frac{f(x,y)}{g(x,y)}]\geq \mbox{Var}_g[\frac{f_1(x)}{g_1(x)}]$

30 Views Asked by At

Prove that $$\mbox{Var}_g[\frac{f(x,y)}{g(x,y)}]\geq \mbox{Var}_g[\frac{f_1(x)}{g_1(x)}]$$ where $$f_1(x) = \int f(x,y) \mathrm{d}y$$ and $f$ and $g$ are distributions.

Here's my attempt:

the LHS is

$$\int \int \frac{f^2}{g^2}g\mathrm{d}x\mathrm{d}y - (\int \int \frac{f}{g}g\mathrm{d}x\mathrm{d}y)^2$$ $$=\int \int \frac{f^2}{g}\mathrm{d}x\mathrm{d}y - (\int \int f\mathrm{d}x\mathrm{d}y)^2$$ $$=\int \int \frac{f^2}{g}\mathrm{d}x\mathrm{d}y - 1.$$

The RHS is

$$\int \int \frac{f_1^2}{g_1^2}g\mathrm{d}x\mathrm{d}y - (\int \int \frac{f_1}{g_1}g\mathrm{d}x\mathrm{d}y)^2$$

$$=\int \frac{f_1^2}{g_1^2}g_1\mathrm{d}x - (\int \frac{f_1}{g_1}g_1\mathrm{d}x)^2$$ $$=\int \frac{f_1^2}{g_1^2}g_1\mathrm{d}x - (\int f_1\mathrm{d}x)^2$$

$$=\int \frac{f_1^2}{g_1}\mathrm{d}x - 1$$

Is there any reason why

$$\int \int \frac{f^2}{g}\mathrm{d}x\mathrm{d}y$$

should have to be greater or equal to

$$\int \frac{f_1^2}{g_1}\mathrm{d}x$$?

1

There are 1 best solutions below

0
On BEST ANSWER

From Cauchy Schwarz inequality,

$$(\int\dfrac{f^2(x,y)}{g(x,y)}dy)(\int g(x,y)dy)\geq (\int f(x,y)dy)^2$$

This yields

$$\int \dfrac{f^2(x,y)}{g(x,y)}dy\geq\dfrac{f_1^2(x)}{g_1(x)}$$

Now integrate both sides w.r.t. $x$ to get your inequality.