虽然文中详细的说明了利用Jensen不等式证明。但是可能有的同学还是不清楚。下面给出推导过程:
已知 :
f(E[x])f(E[q(x)p(x)])≤E[f(x)]≤E[f(q(x)p(x))]
其中 f(x)=−lnx
则:
f(E[q(x)p(x)])−ln∫p(x)q(x)p(x)dx−ln∫q(x)dx−ln10≤E[f(q(x)p(x))]≤∫p(x)⋅−lnq(x)p(x)dx≤−∫p(x)lnq(x)p(x)dx≤−∫p(x)lnq(x)p(x)dx≤−∫p(x)lnq(x)p(x)dx