A. How do I prove the following sequence converges as $n$ goes to $\infty$ for any $c$, and how do I find the limit?

$$ \begin{align} a_1 &=\frac{1}{c} \int _0^c\frac{x_1 }{1+x_1}\;dx_1 \\ \\ a_2 & =\frac{1}{c^2 } \int _0^c\int _0^c\frac{x_1 +x_2 }{2+x_1 +x_2 }\;dx_2\;dx_1 \\ \\ a_3 & =\frac{1}{c^3 } \int _0^c\int _0^c\int _0^c\frac{x_1 +x_2 +x_3 }{3+x_1 +x_2 +x_3 } \;dx_3\;dx_2\;dx_1 \end{align} $$

and so on for $a_n$ $\dots$

B. Similarly, with this, where $f$ and $g$ are not polynomials [I verified the convergence numerically]:

$$ \begin{align} a_1 &=\frac{1}{c} \int _0^c\frac{f(x_1) }{g(x_1)}\;dx_1 \\ \\ a_2 & =\frac{1}{c^2 } \int _0^c\int _0^c\frac{f(x_1) +f(x_2) }{g(x_1) +g(x_2) }\;dx_2\;dx_1 \\ \\ a_3 & =\frac{1}{c^3 } \int _0^c\int _0^c\int _0^c\frac{f(x_1) +f(x_2) +f(x_3) }{g(x_1) +g(x_2) +g(x_3) } \;dx_3\;dx_2\;dx_1 \end{align} $$

and so on for $a_n$ $\dots$ (perhaps I need to include the very long definition of $f$ and $g$...?)


Let $S_n$ denote the sum of $n$ i.i.d. random variables uniformly distributed on $(0,c)$. Then $a_n=\mathrm E\left(\dfrac{S_n}{n+S_n}\right)$. By the strong law of large numbers, $\dfrac{S_n}n\to\dfrac{c}2$ almost surely, hence $$\lim\limits_{n\to\infty} a_n=\dfrac{c/2}{1+c/2}=\dfrac{c}{2+c}.$$

Edit In case B (added afterwards to the question) one should consider the sums $S_n$ and $T_n$ of $n$ random variables $f(X_i)$ and $g(X_i)$, for some i.i.d. random variables $(X_n)_{n\geqslant1}$ uniformly distributed on $(0,c)$. Then $a_n=\mathrm E(R_n)$ with $R_n=S_n/T_n$ and, by the strong law of large numbers, $S_n/n\to\mathrm E(f(X_1))$ and $T_n/n\to\mathrm E(g(X_1))$ hence $R_n\to\mathrm E(f(X_1))/\mathrm E(g(X_1))$ almost surely.

If, for example, $g\gt0$ almost everywhere on $(0,c)$, $R_n$ is well defined. If, for example, $|f|\leqslant Ag$ almost everywhere for some finite $A$, $(R_n)_{n\geqslant1}$ is uniformly integrable hence $$\lim\limits_{n\to\infty} a_n=\lim\limits_{n\to\infty}\mathrm E(R_n)=\mathrm E(\lim\limits_{n\to\infty} R_n)=\dfrac{\mathrm E(f(X_1))}{\mathrm E(g(X_1))}=\dfrac{\displaystyle\int_0^cf(x)\mathrm dx}{\displaystyle\int_0^cg(x)\mathrm dx}.$$

  • $\begingroup$ Thanks, Didier. This was the 1st part of my question. The 2nd part involves replacing the x_i with f(x_i) in the numerator and g(x_i) in the denominator (and removing the constant), where f and g are not polynomials. They are rather complicated functions so I am not including their definition here. However, numerical simulations demonstrate the convergence. I add this now to the main question above. $\endgroup$ – Omri Jan 16 '12 at 22:43
  • $\begingroup$ Out of personal curiosity and perhaps for the enhancement of this question's usefulness, could you provide those functions and their context? You do not have to, this is simply a suggestion. $\endgroup$ – 000 Jan 16 '12 at 23:06
  • $\begingroup$ Omri: Do not do that (adding new questions to a post). $\endgroup$ – Did Jan 17 '12 at 6:33
  • $\begingroup$ OK Didier, and thanks a lot for the continued answer... $\endgroup$ – Omri Jan 17 '12 at 7:39
  • $\begingroup$ Didier, one last point: if the n i.i.d random variables are not uniformly distributed (actually, a Beta), so that the integrand is multiplied by the corresponding pdf (and a proper normalization replaces c^n), does your final limit still apply? $\endgroup$ – Omri Jan 17 '12 at 10:23

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.