1

Just wondering if there is any other way to show that for each positive integer $n$ holds

$$2\left(\sqrt{n} - 1\right) < 1 + \frac{1}{\sqrt{2}} +\cdots+\frac{1}{\sqrt{n}} < 2\sqrt{n}$$

other than by mathematical induction~

mechanodroid
  • 46,490
Homaniac
  • 1,215

6 Answers6

2

For the right inequality:

$$\frac{1}{\sqrt{n}}\left(1+\frac{1}{\sqrt{2}}+\cdots+\frac{1}{\sqrt{n}}\right) = \frac{1}{n}\left(\sqrt{n}+\sqrt{\frac{n}{2}}+\cdots+\sqrt{\frac{n-1}{n}} + \sqrt{\frac{n}{n}}\right)$$

This is a Riemann sum of a function $f(x) = \frac{1}{\sqrt{x}}$ on the segment $[0,1]$ with the equidistant subdivision with endpoints $\left\{0, \frac{1}{n}, \frac{2}{n}, \ldots, 1\right\}$.

So:

$$\lim_{n\to\infty} \frac{1}{n}\left(\sqrt{n}+\sqrt{\frac{n}{2}}+\cdots+ \sqrt{\frac{n}{n}}\right) = \int_{0}^1 \frac{dx}{\sqrt{x}} = 2$$

Since the Riemann sums increase towards the integral, by multiplying with $\sqrt{n}$ we obtain the desired inequality.

mechanodroid
  • 46,490
2

$f(x)=\frac{1}{\sqrt{x}}$ is a convex function on $\mathbb{R}^+$, hence by the Hermite-Hadamard inequality

$$ \frac{1}{\sqrt{1}}+\ldots+\frac{1}{\sqrt{n}}\leq\int_{1/2}^{n+1/2}\frac{dx}{\sqrt{x}}=\sqrt{4n+2}-\sqrt{2}\leq 2\sqrt{n}-1 $$ as well as

$$ \frac{1}{\sqrt{1}}+\ldots+\frac{1}{\sqrt{n}}\geq \frac{1}{2\sqrt{1}}+\frac{1}{2\sqrt{n}}+\int_{1}^{n}\frac{dx}{\sqrt{x}} \geq 2\sqrt{n}-\tfrac{3}{2}. $$

Jack D'Aurizio
  • 353,855
1

Note that $$\frac{1}{k+1}<\int_k^{k+1}\frac{dx}{\sqrt{x}}<\frac{1}{k}$$ for every $k\geq0$. Adding this relations we get $$\int_1^{n+1}\frac{dx}{\sqrt{x}}<1 + \frac{1}{\sqrt{2}} +\cdots+\frac{1}{\sqrt{n}}<\int_0^n\frac{dx}{\sqrt{x}}$$ Solving the integrals we get what we wanted.

Veridian Dynamics
  • 2,471
  • 9
  • 14
1

Consider the graph of $f(x)=\frac{1}{\sqrt{x}}$. Observe that this is a decreasing function.

For the upper bound: Divide up the interval $[0,n]$ into $n$ subintervals of size $1$. Now, consider the rectangles of the form $[a,a+1]\times[0,\frac{1}{\sqrt{a+1}}]$. The area of each rectangle is $\frac{1}{\sqrt{a+1}}$. Therefore, the sum of the areas of these rectangles is exactly the sum in the middle. On the other hand, the area of these rectangles is less than the area under the curve $f(x)$. Since $$ \int_0^n\frac{dx}{\sqrt{x}}=\left.2\sqrt{x}\right|_0^n=2\sqrt{n}, $$ the result follows.

For the lower bound: This is similar, but now we want to compare to $\int_1^n\frac{dx}{\sqrt{x}}$. In this case, one would divide up the interval $[1,n+1]$ into $n$ subintervals of size $1$. Now, consider the rectangles of the form $[a,a+1]\times[0,\frac{1}{\sqrt{a}}]$ the area of each rectangle is $\frac{1}{\sqrt{a}}$. Therefore the sum of the areas of these rectangles is exactly the sum on the middle. On the other hand, the area of these rectangles is greater than the area under the curve $f(x)$. Since $$ \int_1^{n+1}\frac{dx}{\sqrt{x}}=2\sqrt{x}|_1^{n+1}=2\sqrt{n+1}-2, $$ and $\sqrt{n}\leq \sqrt{n+1}$, the result follows.

Michael Burr
  • 32,867
  • May i ask in particular for the lower bound: How does the area of each rectangle being root a leads to the sum of these rectangles being the sum in the middle? – Homaniac Sep 07 '17 at 23:04
  • @xiangqi You're right, I forgot the fraction! – Michael Burr Sep 07 '17 at 23:30
  • Why is the total area of rectangles smaller than area under curve for the first case and larger for the second? The difference is we have chosen different intervals to divide into but I can't see how that works.. – Homaniac Sep 07 '17 at 23:55
  • If you draw the picture, in one case the rectangles are under the curve, but in the other, the rectangles are above the curve. Notice that the rectangles are shifted in the two examples. – Michael Burr Sep 08 '17 at 00:34
1

The outcome of the standard sum/integral comparison is $$ 2 \sqrt {n+1} - 2 \; < \; 1 + \frac{1}{\sqrt 2 } + \frac{1}{\sqrt 3 } + \cdots + \frac{1}{\sqrt n } \; < \; 2 \sqrt n - 1 $$

enter image description here

Will Jagy
  • 139,541
0

Upper Bound:

$$H_k=1+\sum_{k=2}^n\frac{1}{n}\leq 1+\sum_{k=2}^n\frac{1}{\sqrt{k(k-1)}}\leq^{CS} 1+\sqrt{(n-1)\sum_{k=2}^n\frac{1}{k(k-1)}}\leq 1+\sqrt{(n-1) \left(\frac{1}{2}-\frac{1}{n}\right)}$$

Which is even better than yours (CS stands for Cauchy-Schwarz).

For the lower bound, work on:

$$\sqrt{k-1}-\sqrt{k}=\frac{1}{\sqrt{k+1}+\sqrt{k}}\approx \frac{2}{\sqrt{k}}$$