9

Let $x$ and $y$ be nonnegative real numbers such that $x+y=1$. How do I show that $\sqrt{x^2+1}+\sqrt{y^2+1}\ge \sqrt{5}$?

How do I deal with square roots inside the inequality?

  • Thank you, Michael. I'm leaving the accepted answer open for now as I am also looking for a pre-calculus solution. –  Aug 14 '13 at 07:46
  • 4
    Fair enough. In future you should make such requests in the question. – Michael Albanese Aug 14 '13 at 07:55
  • Isn't this fairly simple? Take vectors $(x,1)$ and $(y,1)$, and apply the triangle inequality. (Or is that how you came up with this?) – Chris Sep 08 '13 at 19:19

7 Answers7

44

Triangle inequality :) The length of the kinked red line represents the sum of the two roots, which is not shorter that the long diagonal of this $1:2$ rectangle.

enter image description here

(Edit: changed my poorly drawn diagram; was this)

peterwhy
  • 22,256
  • 2
    Very nice. I was just writing up an answer similar to this. – Michael Albanese Aug 14 '13 at 08:04
  • 5
    Very nice! If anyone is struggling to see why this is proof, think what the shortest length possible for the kinked red line is. The black line across the middle of the picture is split into two lengths x and y which sum to 1. Where this division occurs is where the red lines meet. The shortest that the sum of the lengths these lines can be is thus the diagonal, sqrt 5! – JH92 Aug 14 '13 at 08:33
  • @peterwhy great answer! – Sepideh Bakhoda Aug 14 '13 at 09:14
14

Assuming the contrary, for some $0\le x \le 1$, and $y=1-x$,

$$\begin{align} \sqrt{x^2+1}+\sqrt{y^2+1} <& \sqrt{5}\\ \left( \sqrt{x^2+1} + \sqrt{y^2+1} \right)^2 <& 5 &\text{(Square both sides)}\\ x^2 +1 + 2\sqrt{ \left( x^2 + 1 \right)\left( y^2 + 1 \right)} + y^2 + 1 <& 5\\ 2\sqrt{ \left( x^2 + 1 \right)\left( y^2 + 1 \right)} <& 3 - x^2 -y^2\\ 2\sqrt{ \left( x^2 + 1 \right)\left( y^2 + 1 \right)} <& 3 - \left(x+y\right)^2 + 2xy\\ 2\sqrt{ \left( x^2 + 1 \right)\left( y^2 + 1 \right)} <& 3 - 1 + 2xy &\text{(Substitute }x+y=1\text{)}\\ \sqrt{\left( x^2+1 \right) \left( y^2 + 1 \right)} <& 1+xy\\ \left( x^2+1 \right) \left( y^2 + 1 \right) <& \left( 1+xy \right)^2 &\text{(Square both sides)}\\ x^2 y^2 + x^2 + y^2 + 1 <& 1 + 2xy + x^2 y^2\\ x^2 -2xy+y^2 <& 0\\ \left( x - y \right)^2 <& 0\\ \end{align}$$

which contradicts.

peterwhy
  • 22,256
7

Let $f(x) = \sqrt{x^2+1}+\sqrt{(1-x)^2+1}$. As $x$ and $y$ are non-negative integers, we only consider $x \in [0, 1]$.

Note that $f(x)$ is symmetric about $x = \frac{1}{2}$ and

$$f'(x) = \frac{x\sqrt{(1-x)^2+1} + (x - 1)\sqrt{x^2+1}}{\sqrt{x^2+1}\sqrt{(1-x)^2+1}}.$$

For $0 < x < \frac{1}{2}$, $x < 1 - x$ so $\sqrt{x^2+1} > \sqrt{(1-x)^2+1}$. Therefore, $$f'(x) = \frac{x\sqrt{(1-x)^2+1} + (x - 1)\sqrt{x^2+1}}{\sqrt{x^2+1}\sqrt{(1-x)^2+1}} < \frac{\frac{1}{2}\sqrt{(1-x)^2+1} - \frac{1}{2}\sqrt{x^2+1}}{\sqrt{x^2+1}\sqrt{(1-x)^2+1}} < 0.$$ So $f$ is decreasing on $[0, \frac{1}{2})$ and as $f$ is symmetric about $x = \frac{1}{2}$, $f$ is increasing on $(\frac{1}{2}, 1]$. Therefore $f$ attains its minimum value at $x = \frac{1}{2}$ which is $$f\left(\frac{1}{2}\right) = \sqrt{\left(\frac{1}{2}\right)^2+1} + \sqrt{\left(1-\frac{1}{2}\right)^2+1} = \sqrt{\frac{5}{4}} + \sqrt{\frac{5}{4}} = \frac{1}{2}\sqrt{5} + \frac{1}{2}\sqrt{5} = \sqrt{5}.$$ Hence, $f(x) \geq \sqrt{5}$ for $x \in [0, 1]$. Setting $y = 1 -x$ we have $\sqrt{x^2+1} + \sqrt{y^2+1} \geq \sqrt{5}$.

6

If you are familiar with Minkowski's inequality, this is straightforward.

$$\sqrt{x^2+1}+\sqrt{y^2+1} \ge \sqrt{(x+y)^2 + (1+1)^2} = \sqrt5$$

Macavity
  • 46,381
  • Really, this is just the triangle inequality for vectors $(x,1)$ and $(y,1)$. (+1, however) – Chris Sep 08 '13 at 19:17
  • With the added comment of user1296727...this is the best answer present here...I really like this... – Hawk Apr 27 '14 at 16:34
1

Another interpretation of the problem is this: $ \ y = \sqrt{x^2 + 1} \ \ \text{and} \ \ x = \sqrt{y^2 + 1} \ $ are the "positive" branches of the "vertical" and "horizontal" unit hyperbolas centered on the origin. They thus share the asymptotes $ \ y = \pm \ x \ $ .

It is specified that $ \ x \ge 0 \ \ \text{and} \ \ y \ge 0 \ $ and that they be related through $ \ x + y = 1 \ . $ So we in fact require that $ \ 0 \ \le \ x,y \ \le \ 1 \ . $ We can thus imagine a point in the first quadrant which slides along the line $ \ y = 1 - x \ $ in said quadrant and consider the lengths of the vertical and horizontal line segments extending from each coordinate axis to the appropriate hyperbolic branch, as depicted in the figure below.

enter image description here

It is clear that the behavior of this geometric arrangment is symmetrical about the asymptote $ \ y = x \ , $ so we only need look at the results for, say, $ \ 0 \ \le \ x \ \le \ \frac{1}{2} \ . $ For $ \ x = 0 \ , $ we attain the maximum for the sum sought,

$$ \sqrt{0^2 + 1} \ + \ \sqrt{1^2 + 1} \ = \ 1 + \sqrt{2} \ \approx \ 2.4142 \ , $$

while for $ \ x = \frac{1}{2} \ , $ the sum has its minimum,

$$ \sqrt{(\frac{1}{2})^2 + 1} \ + \ \sqrt{(\frac{1}{2})^2 + 1} \ = \ 2 \cdot \sqrt{\frac{5}{4}} \ = \ \sqrt{5} \ \approx \ 2.2361 \ . $$

The sum of radicals has a fairly narrow range of values for the specified domain. [This approach is most nearly similar to Michael Albanese's analysis. (It also seems a sort of "inside-out" version of peterwhy's elegant method.)]

colormegone
  • 10,842
0

Construct a right-angled triangle, $1$ unit and $2$ units long for two perpendicular sides. The length of hypotenuse is? And note that straight line gives the shortest distance between two points.

Michael Li
  • 2,201
0

Single variable calculus.

$x+y=1 \implies? \sqrt{x^2+1}+\sqrt{y^2+1} \ge \sqrt{5}$

$y=1-x$.

$\sqrt{x^2+1}+\sqrt{(1-x)^2+1}=\sqrt{x^2+1}+\sqrt{x^2-2x+2}$

$\frac{x}{\sqrt{x^2+1}}+\frac{x-1}{\sqrt{x^2-2x+2}}=0$

$x\sqrt{x^2-2x+2}=-(x-1)\sqrt{x^2+1}$

$x^2(x^2-2x+2)=x^4-2x^3+2x^2=(x^2+1)(x^2-2x+1)=x^4-2x^3+x^2+x^2-2x+1$

$\implies x=1/2, y=1/2$


Lagrange Multipliers.

$g(x,y)=x+y-1=0$

$f(x,y)=\sqrt{x^2+1}+\sqrt{y^2+1}$

$\frac{2x}{\sqrt{x^2+1}}=\lambda$

$\frac{2y}{\sqrt{y^2+1}}= \lambda$

$\frac{x}{\sqrt{x^2+1}}=\frac{y}{\sqrt{1+y^2}}$

$\frac{x^2}{x^2+1}=\frac{y^2}{1+y^2}$

$x^2+x^2y^2=y^2x^2+y^2\implies x^2=y^2\implies x=y=1/2$

So the minimum is $\sqrt{5}$.

This method can be generalized. If your constraint function and objective function are symmetric under exchange of theire variables, i.e. $f(x,y)=f(y,x), g(x,y)=g(y,x)$, extreme values typically obtain when $x=y$.


Rule of thumb suggests we start with $x=y=1/2$ given the symmetry of the problem. Then demonstrate deviations shift away from the extrema.

Suppose $x=(1/2)+p$ and $y=(1/2)+q$

Since $x+y=1$ we must have $q=-p$

$u=\sqrt{1+x^2}+\sqrt{1+y^2}=\sqrt{5/4+p^2+p}+\sqrt{5/4+p^2-p}$

$u^2=5/2+2p^2+2\sqrt{25/16+p^4+3p^2/2}$

From here its clear $p=0$ minimizes the expression.

TurlocTheRed
  • 5,683