Let $X_A$ and $X_B$ be independent exponentially distributed RVs with means $\mu_A$ and $\mu_B$, respectively.

I have been running simulations that suggest $$\mathbb E \left[ \frac{X_B}{A X_A+BX_B+C} \right] \leq \mathbb E \left[ \frac{X_B}{A \mu_A+B\mu_B+C} \right],$$ where $A,B,C>0$ are constants.

My question is, does anyone no how I might go about proving this?


If this is true for every positive $(A,B,C)$, then, by continuity when $B\to0$, $$\mathbb E \left[ \frac{X_B}{A X_A+C} \right] \leqslant \mathbb E \left[ \frac{X_B}{A \mu_A+C} \right]=\frac{\mu_B}{A \mu_A+C}.$$ By independence and strict convexity of the function $x\mapsto1/(Ax+C)$ on $x\gt0$, the LHS is $$\mu_B\mathbb E \left[ \frac{1}{A X_A+C} \right]\gt\mu_B\frac{1}{A \mathbb E(X_A)+C}=\frac{\mu_B}{A\mu_A+C}.$$ Thus, the inequality cannot hold in full generality.

On the other hand, if $A\mu_A=B\mu_B$, then $AX_A$ and $BX_B$ are i.i.d. hence, by symmetry, $$2B\,\mathbb E \left[ \frac{X_B}{A X_A+BX_B+C} \right]=\mathbb E \left[ \frac{AX_A+BX_B}{A X_A+BX_B+C} \right]=1-C\,\mathbb E \left[ \frac{1}{A X_A+BX_B+C} \right],$$ and, by convexity, $$\mathbb E \left[ \frac{1}{A X_A+BX_B+C} \right]\gt\frac{1}{\mathbb E(A X_A+BX_B)+C}=\frac{1}{A\mu_A+B\mu_B+C},$$ hence, indeed, $$\mathbb E \left[ \frac{X_B}{A X_A+BX_B+C} \right]\leqslant\frac{\mu_B}{A\mu_A+B\mu_B+C}.$$

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.