From the definition of a definite integral as the limit of a sum, evaluate $$\int_a^b\frac{1}{\sqrt x}dx$$

My progress so far:

$\begin{align} \int_a^b\frac{1}{\sqrt x}dx & =\lim_{n\to\infty}h\times\bigg(f(a)+f(a+h)+\cdots+f\big(a+(n-1)h\big)\bigg) \\ & = \lim_{n\to\infty}h\times\bigg(\frac1{\sqrt a}+\frac1{\sqrt {a+h}} +\cdots+\frac1{\sqrt {a+(n-1)h}}\bigg) \end{align}$

After this point, I cant't think of a way to combine the reciprocals of square roots into a neat form. Usually a convenient summation formula is applicable but how would I proceed in this case?

  • $\begingroup$ Are you sure you're not given some sort of constants $a,b$? I don't recall ever seeing one without them in my school work, seems like it might make things a little too difficult at first glance. $\endgroup$ – Eevee Trainer Mar 29 at 2:45
  • $\begingroup$ @EeveeTrainer Yes, all the other questions in the exercise also have limits $a$,$b$. Using $nh=b-a$ still works even if you don't know what $a$ and $b$ are. $\endgroup$ – s0ulr3aper07 Mar 29 at 2:49

We assume $0\leq a\leq b$. In order to cope with the square root function $\frac{1}{\sqrt{x}}$ it is convenient to use variable length intervals with length $j^2\frac{b-a}{n^2}$. When taking square roots we can factor out $j$ and summation is expected to become simpler. The corresponding Riemann sum is \begin{align*} \int_{a}^{b}\frac{1}{\sqrt{x}}\,dx=\lim_{n\to \infty}\sum_{j=1}^n\frac{1}{\sqrt{\color{blue}{a}+ j^2\frac{b-a}{n^2}}}\cdot\left(j^2\frac{b-a}{n^2}-(j-1)^2\frac{b-a}{n^2}\right)\tag{1} \end{align*} But it's still not easy due to the constant $\color{blue}{a}$. To overcome this difficulty we write the integral as difference of two improper integrals \begin{align*} \int_{a}^{b}\frac{1}{\sqrt{x}}\,dx=\int_{0}^{b}\frac{1}{\sqrt{x}}\,dx-\int_{0}^{a}\frac{1}{\sqrt{x}}\,dx\tag{2} \end{align*} If the limit of the Riemann sums of the improper integrals exists we have found the wanted Riemann integral and we are done.

We subdivide the interval $[0,b]$ in \begin{align*} [0,b]=\bigcup_{j=1}^n\left[(j-1)^2\frac{b}{n^2},j^2\frac{b}{n^2}\right] \end{align*} with length \begin{align*} j^2\frac{b}{n^2}-(j-1)^2\frac{b}{n^2}=\frac{(2j-1)b}{n^2}\qquad\qquad1\leq j\leq n \end{align*} and obtain the Riemann sum \begin{align*} \color{blue}{\int_{0}^b\frac{1}{\sqrt{x}}\,dx}&=\lim_{n\to\infty}\sum_{j=1}^n\frac{1}{\sqrt{\frac{j^2b}{n^2}}}\cdot\frac{(2j-1)b}{n^2}\tag{3}\\ &=\sqrt{b}\lim_{n\to\infty}\frac{1}{n}\sum_{j=1}^n\left(2-\frac{1}{j}\right)\tag{4}\\ &=\sqrt{b}\lim_{n\to\infty}\frac{1}{n}\left(2n-H_n\right)\tag{5}\\ &\,\,\color{blue}{=2\sqrt{b}}\tag{6} \end{align*}

We finally conclude from (2) and (6) \begin{align*} \color{blue}{\int_{a}^b\frac{1}{\sqrt{x}}\,dx=2\left(\sqrt{b}-\sqrt{a}\right)} \end{align*}


  • In (3) we have the now convenient representation (1) evaluated at $a=0$.

  • In (4) we do some simplifications and factor out terms not dependent on $n$ resp. $j$.

  • In (5) we note the Harmonic numbers $H_n=\sum_{j=1}^n \frac{1}{j}$ grow asymptotically with $\log n$, so that \begin{align*} \lim_{n\to\infty}\frac{H_n}{n}=\lim_{n\to\infty}\frac{\log n}{n}=0. \end{align*}

  • $\begingroup$ Thanks for the beautiful solution. However, while using $log \ n$ to approximate $H_n=\sum_{j=1}^n \frac{1}{j}$, there is difference of exactly $\gamma = 0.57721\ldots$ between the two when $\lim_{n\to \infty}$. So maybe adding the constant value would make this already brilliant answer even better? $\endgroup$ – s0ulr3aper07 Mar 29 at 18:24
  • $\begingroup$ Also, $j^2\frac{b}{n^2}-(j-1)^2\frac{b}{n^2}=\frac{(2j-1)b}{n^2}$ is I believe what you intended to write. $\endgroup$ – s0ulr3aper07 Mar 29 at 18:26
  • $\begingroup$ @s0ulr3aper07: You're welcome and many thanks for pointing to the mistakes. $\endgroup$ – Markus Scheuer Mar 29 at 18:39
  • 1
    $\begingroup$ @s0ulr3aper07: With respect to the harmonic numbers I think it's sufficient to use the dominant term $\log n$ of the asymptotic expansion, since we are dividing by $n$ when taking the limit. $\endgroup$ – Markus Scheuer Mar 29 at 18:40
  • $\begingroup$ I tried a similar approach for $\int_a^b{\frac1{x^2}dx}$ with variable length intervals of length $\sqrt{j} \frac{b-a}{\sqrt{n}}$ but unlike the resulting calculations above, these new calculations are not as clean. Specially because I get $\sum_{j=1}^n \frac{1}{j}(\sqrt{j}-\sqrt{j-1}$). Could you provide some help there please? $\endgroup$ – s0ulr3aper07 Mar 30 at 8:09

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.