I know that $$\Delta x=\frac{b-a}{n}=\frac{1-0}{n}=\frac{1}{n}$$ So $$\overline{D}=\displaystyle\lim_{n\to\infty}\displaystyle\sum_{i=1}^n\sup f(x_i)\Delta x=\displaystyle\lim_{n\to\infty}\displaystyle\sum_{i=1}^nf(x_i)\Delta x=\displaystyle\lim_{n\to\infty}\displaystyle\sum_{i=1}^n \frac{1}{n}\sqrt {\frac{i}{n}}=\displaystyle\lim_{n\to\infty}\frac{1}{n\sqrt n}\displaystyle\sum_{i=1}^n \sqrt {i} $$ I don't know how to calculate the last part $\displaystyle\sum_{i=1}^n \sqrt {i}$
-
2In the limit, it should be $n\to\infty$ instead of $i\to\infty$ – 5xum Apr 06 '16 at 10:50
-
The Cauchy-Schwarz inequality yields $$\sum_{i=1}^n i^{\frac12} \leqslant \sqrt n\sum_{i=1}^n i. $$ – Math1000 Apr 06 '16 at 10:53
-
There is no simple formula, or derivation. One usually goes the other way - knowing the integral, one deduces the (asympotic) value of the sum. http://math.stackexchange.com/questions/1241864/sum-of-square-roots-formula – leonbloy Apr 06 '16 at 11:11
2 Answers
It is possible to compute the area (or integral) directly as the limit of the upper Darboux sum.
As you showed
$$\overline{D} = \lim_{n\to\infty}\frac{1}{n\sqrt n} \sum_{i=1}^n \sqrt {i}.$$
We can use the binomial expansion $(1-x)^{\alpha} =1 - \alpha x + O(x^2).$
For $i > 1,$ we have
$$\begin{align}(i-1)^{3/2} &= i^{3/2}(1 - i^{-1})^{3/2} \\ &= i^{3/2}\left[ 1 - \frac{3}{2}i^{-1} + O(i^{-2})\right] \\ &= i^{3/2} - \frac{3}{2}\sqrt{i} + O(1/\sqrt{i}).\end{align}$$
Hence,
$$\sqrt{i} = \frac{2}{3}\left[i^{3/2} - (i-1)^{3/2}\right] + O(1/\sqrt{i}),$$
and,
$$\begin{align}\frac1{n\sqrt{n}}\sum_{i=1}^n\sqrt{i} &=\frac{1}{n \sqrt{n}}\left(1 + \frac{2}{3}\sum_{i=2}^n[i^{3/2}-(i-1)^{3/2} + O(1/\sqrt{i})]\right) \\ &= \frac{1}{n \sqrt{n}}\left(1 + \frac{2}{3}n^{3/2}- \frac{2}{3} + O(n)\right) \\ &= \frac{2}{3} +\frac{1}{3n \sqrt{n}} + O(1/\sqrt{n}). \end{align}$$
Thus,
$$\lim_{n \to \infty}\frac1{n \sqrt{n}}\sum_{i=1}^n\sqrt{i} = \frac{2}{3}.$$

- 90,707
-
This one is good +1. Using $(x + 1)^{n} - x^{n} = nx^{n - 1} + \cdots$ even when $n$ is not a positive integer is nice but needs analysis of the error terms which is handled very well here. – Paramanand Singh May 19 '16 at 09:09
Let $f$ be bounded on closed interval $[a, b]$. Let $P = \{x_{0} = a, x_{1}, x_{2}, \ldots, x_{n} = b\}$ be a partition of $[a, b]$ then the upper Darboux sum for $f$ over $P$ is given by $$U(P, f) = \sum_{i = 1}^{n}M_{i}(x_{i} - x_{i - 1})\tag{1}$$ and the lower Darboux sum of $f$ over $P$ is given by $$L(P, f) = \sum_{i = 1}^{n}m_{i}(x_{i} - x_{i - 1})\tag{2}$$ where $$M_{i} = \sup\, \{f(x) \mid x \in [x_{i - 1}, x_{i}]\},\, m_{i} = \inf\,\{f(x)\mid x \in [x_{i - 1}, x_{i}]\}\tag{3}$$ Now it is much easier to calculate the area under $f(x) = x^{p}$ for general $p > 0$ on interval $[a, b]$ via limit of Darboux sums. And then we can put $p = 1/2, a = 0, b = 1$ to get the answer to the current question.
For this question it is better to assume that $0 < a < b$ let the partition $P$ of $[a, b]$ be given by $x_{i} = ar^{i}$ when $b = x_{n} = ar^{n}$ and as $n \to \infty$ we have $r \to 1$. Thus the points of partition are in geometric progression (instead of the usual points $x_{i} = a + ih, b = x_{n} = a + nh$ in arithmetic progression). Then since $f(x) = x^{p}$ is increasing we have $m_{i} = x_{i - 1}^{p}, M_{i} = x_{i}^{p}$ and therefore \begin{align} U(P, f) &= \sum_{i = 1}^{n}x_{i}^{p}(x_{i} - x_{i - 1})\notag\\ &= \sum_{i = 1}^{n}a^{p}r^{ip}(ar^{i} - ar^{i - 1})\notag\\ &= a^{p + 1}(r - 1)\sum_{i = 1}^{n}r^{ip + i - 1}\notag\\ &= a^{p + 1}(r - 1)r^{p}\frac{r^{(p + 1)n} - 1}{r^{p + 1} - 1}\notag\\ &= a^{p + 1}r^{p}(r^{(p + 1)n} - 1)\frac{r - 1}{r^{p + 1} - 1}\notag\\ &= a^{p + 1}r^{p}((b/a)^{p + 1} - 1)\frac{r - 1}{r^{p + 1} - 1}\notag\\ \end{align} This tends to $$a^{p + 1}((b/a)^{p + 1} - 1)\frac{1}{p + 1} = \frac{b^{p + 1} - a^{p + 1}}{p + 1}$$ as $r \to 1$. It can be proved in similar manner that $L(P, f)$ also tends to same limit. Hence we have $$\int_{a}^{b}x^{p}\,dx = \frac{b^{p + 1} - a^{p + 1}}{p + 1}\tag{4}$$ Note that in the above we have assumed $p > 0, b > a > 0$. The same result holds if $-1 < p \leq 0$ (only values of $m_{i}$ and $M_{i}$ are changed). Also it is possible to use the result in this question to show that the formula $(4)$ is valid for $0 \leq a < b$ also.
Now putting $p = 1/2, a = 0, b = 1$ in formula $(4)$ we get the desired area under graph of $f(x) = \sqrt{x}$ between $x = 0$ and $x = 1$ as $$\int_{0}^{1}\sqrt{x}\,dx = \frac{1^{3/2} - 0^{3/2}}{3/2} = \frac{2}{3}$$

- 87,309
-
Good to show non-uniform partition as an alternative +1. I think it's an instructive exercise to work such limits directly. Such questions as $\lim \sum_{k=1}^n (n+k)^{-1}$ come up frequently on this site. I find it somewhat amusing how the answer that declares this to be a Riemann sum is typically upvoted massively. – RRL May 19 '16 at 17:32
-
@RRL: Good point. That a limit can be recognized as a limit of Riemann sum does not automatically help in evaluating it. The evaluation is made easy by the use of fundamental theorem of calculus which allows the limit to be expressed in terms of anti-derivative and such answers never emphasize the value of FTC in such a limit evaluation. May be it is obvious to the person who gave the answer but perhaps OP may not be aware of the miracle being performed via FTC. – Paramanand Singh May 19 '16 at 19:34