Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Kullback in his "Information theory and statistics" gives the symmetrized divergence as follows

$J(1,2)=\iint(f(x,y)-g(x)h(y))log{\frac{f(x,y)}{g(x)h(y)}}$

Later (p.8), he states that symmetrized divergence between a bivariate normal distribution $f(x,y)=N(\boldsymbol{\mu,\boldsymbol{\Sigma}})$ and two univariate normal distributions$g(x),h(y)=N(0,\sigma)$. (i.e. means for all distributions are zero)

is $J(1,2)=\frac{\rho^2}{1-\rho^2}$.

I proved that the divergence $I(f(x,y):g(x),h(y))=-\frac{1}{2} log{(1-\rho^2)}$ and I can see that $I(g(x),h(y):f(x,y))$ has to be $\frac{1}{2} log{(1-\rho^2)}+\frac{\rho^2}{1-\rho^2}$ but I can't prove the second part.

$I(g(x),h(y):f(x,y))=\frac{1}{2} log{(1-\rho^2)}+\iint{\{g(x)h(y)-\frac{1}{2}[\frac{x^2}{\sigma_x}+\frac{y^2}{\sigma_y}]+\frac{1}{2(1-\rho^2)}[\frac{x^2}{\sigma_x}+\frac{y^2}{\sigma_y}-2\rho\frac{xy}{\sigma_x\sigma_y}]\}dxdy}$

share|improve this question
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.