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.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Suppose that for each $n\geq1$ there is a given sequence of independent random variables $$\xi_{n1},\xi_{n2},\ldots,\xi_{nn}$$ with $E\xi_{nk}=0$, $V\xi_{nk}=\sigma_{nk}^2$, $\sum_{k=1}^n\sigma_{nk}^2=1$. Let $S_n=\xi_{n1}+\ldots+\xi_{nn},$ $$F_{nk}(x)=P\{\xi_{nk}\leq x\},\ \Phi(x)=(2\pi)^{-1/2}\int_{-\infty}^x e^{-y^2/2}dy,\ \Phi_{nk}(x)=\Phi\left(\frac{x}{\sigma_{nk}}\right)$$ On page 339 there is an inequality: $$\sum_{k=1}^n\left| t\int_{-\infty}^{\infty}(e^{itx}-1-itx)(F_{nk}(x)-\Phi_{nk}(x))dx\right|\leq\frac{|t|^3}{2}\varepsilon\sum_{k=1}^n\int_{|x|\leq\varepsilon}|x||F_{nk}(x)-\Phi_{nk}(x)|dx+2t^2\sum_{k=1}^n\int_{|x|>\varepsilon}|x||F_{nk}(x)-\Phi_{nk}(x)|dx$$ How can I get it?

Then the book says we can use $$E|\xi|^n=\int_{-\infty}^{\infty}|x|^ndF(x)=n\int_0^\infty x^{n-1}[1-F(x)+F(-x)]dx$$ to prove $$\int_{|x|\leq\varepsilon}|x||F_{nk}(x)-\Phi_{nk}(x)|dx\leq2\sigma_{nk}^2$$ But I can't figure why.

Thank you!

share|cite|improve this question
up vote 1 down vote accepted

For the second question, rewrite $$ \mathbb E|\xi|^n=\int_{-\infty}^{\infty}|x|^n\mathrm dF(x)=\int_{0}^{\infty}x^n\mathrm dF(x)+ \int_{0}^{\infty}x^n\mathrm dG(x), $$ with $G:x\mapsto-F(-x)$ and use an integration by parts formula to compute $$ \int_0^{+\infty}u(x)\mathrm dv(x),\qquad u:x\mapsto x^n,\quad v=F+G-1.$$ To use this identity, note that for any random variables $\zeta_1$ and $\zeta_2$ with PDF $G_1$ and $G_2$, $$ \int_{-\infty}^{+\infty}|x|\,|G_1(x)-G_2(x)|\mathrm dx\leqslant(*), $$ with $$ (*)=\int_{-\infty}^0(-x)\,(G_1(x)+G_2(x))\mathrm dx+\int_0^{+\infty}x\,(1-G_1(x)+1-G_2(x))\mathrm dx. $$ One sees that $(*)=A(G_1)+A(G_2)$, with $$ A(G)=\int_0^{+\infty}x\,(1-G(x)+G(-x))\mathrm dx. $$ Applying the first identity one gets $A(G)=\frac12\mathbb E(\zeta^2)$ for every random variable $\zeta$ with PDF $G$, hence $(*)=\frac12(\mathbb E(\zeta_1^2)+\mathbb E(\zeta_2^2))$. In particular, the inequality mentioned in the post holds, without the restriction to $|x|\leqslant\varepsilon$ (this was expected) and without the factor $2$ on the RHS (this was less expected).

share|cite|improve this answer
Thank you very much! And how can I use the equation you proved$$E|\xi|^n=\int_{-\infty}^{\infty}|x|^ndF(x)=n\int_0^\infty x^{n-1}[1-F(x)+F(-x)]dx$$to prove$$\int_{|x|\leq\varepsilon}|x||F_{nk}(x)-\Phi_{nk}(x)|dx\leq2\sigma_{nk}^2?‌​$$ – Danielsen Dec 30 '12 at 14:10
Thank you very much! – Danielsen Dec 31 '12 at 1:08

We have that $|e^{itx}-1-itx|\leqslant \frac{|tx|^2}2$, which can be obtained by Taylor's formula, and this gives an upper bound for $\int_{-\varepsilon}^{\varepsilon}$.

For the other part, note that $$|e^{itx}-1-itx|\leqslant|e^{itx}-1|+|tx|\leqslant\left|\int_0^{|tx|}ie^{is}ds\right|+|tx|\leqslant 2|tx|,$$ and we can control $\int_{\{|x|>\varepsilon\}}$.

share|cite|improve this answer
And how about my second question? I still got no idea. Thank you very much! – Danielsen Dec 27 '12 at 9:40

Your Answer


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

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