Given a function $\alpha$ of bounded variation on $[0,1]$, show that: $\lim_{n \rightarrow \infty} \int_{0}^{1} x^n d\alpha(x) =\alpha(1) - \alpha(1-)$. Where $\alpha(1-)$ is the left sided limit.

First off, I can't see how this limit is not $0$, I know it is zero if $\alpha$ is continuous. I'm still trying to find an example of a discontinuous $\alpha$ for which this limit is different than zero.

Now, this is what I've attempted for this limit: Since $\alpha$ is of bounded variation, $\alpha = \beta - \gamma$ for $\beta$ and $\gamma$ increasing functions. So it suffices to show this for an increasing function, say $\beta$.

So, $\int_{0}^{1} x^n d\beta(x) = \int_{0}^{1-\epsilon} x^n d\beta(x) + \int_{1-\epsilon}^{1} x^n d\beta(x)$ for all $\epsilon \in (0,1)$.

Then, $0\le \int_{0}^{\epsilon}x^nd\beta(x) \le \epsilon^n(\beta(\epsilon) - \beta(0))$, so that $\lim_{n \rightarrow \infty} \int_{0}^{1-\epsilon} x^n d\beta(x) = 0$. This means that $\lim_{n \rightarrow \infty} \int_{1-\epsilon}^{1} x^n d\beta(x)$ should be equal to $\alpha(1) - \alpha(1-)$. But I still don't know how to prove this.

  • $\begingroup$ Why is $$0\le \int_{0}^{\epsilon}x^nd\beta(x) \le \epsilon^n(\beta(\epsilon) - \beta(0))$$ true? If you're using $0 \leq x^n \leq 1$ and monotonicity of integration, then why is it $\epsilon^n\big(\beta(\epsilon)-\beta(0)\big)$ instead of $\epsilon\big(\beta(\epsilon)-\beta(0)\big)$? Why the exponent $n$? $\endgroup$ – stressed out Mar 11 at 17:11
  • $\begingroup$ Take $\alpha(x)=0$ for $x<1$ and $\alpha(1)=1$. Then the answer is just $\alpha(1)-\alpha(1-)=1-0=1$. $\endgroup$ – Alex R. Mar 11 at 17:36
  • $\begingroup$ @stressedout Because by the mean value theorem for integrals there is a value $x_0 \in (0,\epsilon)$ with such that $\int_{0}^{\epsilon}x^nd\beta(x) = x_0^n(\beta(\epsilon) - \beta(0))$ and since $x^n$ is increasing, I could just take $\epsilon^n$ so that $\int_{0}^{\epsilon}x^nd\beta(x) \le \epsilon^n(\beta(\epsilon) - \beta(0))$ because $x_0\leq\epsilon$ and because $\beta$ is increasing. $\endgroup$ – user397146 Mar 11 at 18:56

Among the many ways to prove this, we can use integration by parts to get

$$\int_0^1 x^n \, d\alpha(x) = 1\cdot\alpha(1) - 0\cdot \alpha(0) - \int_0^1 \alpha(x) dx^n = \alpha(1) - n\int_0^1 x^{n-1} \alpha(x) \,dx $$

Changing variables with $x= u^{1/n}$ in the integral on the RHS, we get

$$\int_0^1 x^n \, d\alpha(x) = \alpha(1) - \int_0^1 \alpha(u^{1/n}) \,du $$

Since $\alpha(u^{1/n}) \to \alpha(1-)$ as $n \to \infty$ for $0 < u < 1$ and $\alpha $ is bounded, we have by the dominated convergence theorem

$$\lim_{n \to \infty}\int_0^1 x^n \, d\alpha(x) = \alpha(1) - \alpha(1-) $$

Following your approach

You have already shown that for the non-decreasing function $\beta$ (arising in the Jordan decomposition of $\alpha$)

$$\lim_{n \to \infty} \int_0^{1 - \epsilon}x^n \, d\beta(x) = 0$$

Over the interval $[1-\epsilon, 1]$ we have for $n > 1/\sqrt{\epsilon}$,

$$\int_{1 - 1/n^2}^1 x^n \, d\beta(x) \leqslant \int_{1 - \epsilon}^1 x^n \, d\beta(x) \leqslant \beta(1) - \beta(1-\epsilon)$$


$$\left( 1 - 1/n^2 \right)^n (\beta(1) - \beta(1 - 1/n^2)) \leqslant \int_{1 - \epsilon}^1 x^n \, d\beta(x) \leqslant \beta(1) - \beta(1-\epsilon)$$

Taking the limit as $n \to \infty$ we get

$$\beta(1) - \beta(1-) \leqslant \liminf_{n \to \infty} \int_{1 - \epsilon}^1 x^n \, d\beta(x) \leqslant \limsup_{n \to \infty} \int_{1 - \epsilon}^1 x^n \, d\beta(x) \leqslant \beta(1) - \beta(1 - \epsilon) $$

Since this is true for arbitrarily small $\epsilon$ it follows that

$$\lim_{n \to \infty}\int_0^1 x^n \, d\beta(x) = \beta(1) - \beta(1-)$$

The same argument applies to $\gamma$, and this leads to the desired result.


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.