Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Which is larger? $\sqrt[n]{x+\delta}-\sqrt[n]{x}$ or $\sqrt[n]{x}-\sqrt[n]{x-\delta}$? Algebraic justilation does not help.

share|cite|improve this question
up vote 6 down vote accepted

$\sqrt[n]{x+\delta}-\sqrt[n]{x}\ \boxed{\phantom{A} }\sqrt[n]{x}-\sqrt[n]{x-\delta}$

$\sqrt[n]{x+\delta}+\sqrt[n]{x-\delta}\ \boxed{\phantom{A} }2\sqrt[n]{x}$

$\frac{\sqrt[n]{x+\delta}+\sqrt[n]{x-\delta}}{2}\ \boxed{\phantom{A} }\sqrt[n]{x}$

$\frac{\sqrt[n]{x+\delta}+\sqrt[n]{x-\delta}}{2}\ \boxed{\phantom{A} }\sqrt[n]{\frac{\left(\sqrt[n]{x+\delta}\right)^n+\left(\sqrt[n]{x-\delta}\right)^n}{2}}$

But $\frac{a+b}{2}\leq\sqrt[n]{\frac{a^n+b^n}{2}}$.

$\tiny{\text{some conditions apply}}$

share|cite|improve this answer

The $n$-th root function is concave increasing on $(0,\infty)$. So its rate of change is decreasing. Hence the second quantity is larger.

Edited to add: Alfe has a nice picture that explains this clearly.

share|cite|improve this answer

Let's generalize it even further: For $\alpha>0$ and $x>\delta>0$, we want to decide which is bigger $\left(x+\delta\right)^\alpha-x^\alpha$ or $x^\alpha-\left(x-\delta\right)^\alpha$.

So, let $f(\alpha)=\left(\left(x+\delta\right)^\alpha-x^\alpha\right)-\left(x^\alpha-\left(x-\delta\right)^\alpha\right)$. We want to decide $\text{sign}(f(\alpha))$.


$\begin{align} f(\alpha) &=\left(x+\delta\right)^\alpha+\left(x-\delta\right)^\alpha-2x^\alpha-x^\alpha \\ &=x^\alpha\left(1+\frac{\delta}{x}\right)^\alpha+x^\alpha\left(1-\frac{\delta}{x}\right)^\alpha-2x^\alpha \\ &=x^\alpha\left(\left(1+\frac{\delta}{x}\right)^\alpha+\left(1-\frac{\delta}{x}\right)^\alpha-2\right) \end{align}$

Note that $x^\alpha>0$, so we can divide by it, and $0<\rho:=\frac{\delta}{x}<1$. Therefore we only need to decide the sign of:


We notice that $g(1)=1+\rho+1-\rho-2=0$, and also that $g'(\alpha)=\alpha\left(1+\rho\right)^{\alpha-1}+\alpha\left(1-\rho\right)^{\alpha-1}>0$.

We conclude that $g$ is strictly increasing, thus $\text{sign}(f(\alpha))=\text{sign}(\alpha-1)$, which means:

  • $\alpha<1 \Rightarrow \left(x+\delta\right)^\alpha-x^\alpha<x^\alpha-\left(x-\delta\right)^\alpha$
  • $\alpha=1 \Rightarrow \left(x+\delta\right)^\alpha-x^\alpha=x^\alpha-\left(x-\delta\right)^\alpha$
  • $\alpha>1 \Rightarrow \left(x+\delta\right)^\alpha-x^\alpha>x^\alpha-\left(x-\delta\right)^\alpha$

For your question, if $n>1$ then $\alpha=\frac{1}{n}<1$, and we have $\left(x+\delta\right)^\alpha-x^\alpha<x^\alpha-\left(x-\delta\right)^\alpha$.

share|cite|improve this answer

Hint: Try using the Mean-Value-Theorem for derivatives.

But there is algebraic justification too, note that $$ a-b = \left(\sqrt[n]{a}-\sqrt[n]{b}\right) \times \left\{(\sqrt[n]{a})^{n-1}+(\sqrt[n]{a})^{n-2}\sqrt[n]{b}+\ldots+\sqrt[n]{a}(\sqrt[n]{b})^{n-2}+(\sqrt[n]{b})^{n-1}\right\} $$

share|cite|improve this answer

How about using the fact that $(a-b)(\sum_{i=0}^{n-1}a^{n-1-i}b^i) = a^n - b^n$

share|cite|improve this answer
+1 for mentioning the fact, but how about using it? how does it help? – Arjang Nov 12 '13 at 7:03
@Arjang: as explained by PaulRS above, for both terms $a^n - b^n$ is the same, so just compare which one has bigger $\sum_{i=1}^{n-1}a^{n-1-i}b^i$ – Petite Etincelle Nov 12 '13 at 12:09

We all know that when $n>1$ the curve $x\mapsto y:=x^n$ $\>(x\geq0)$ gets steeper and steeper as $x$ increases. It follows that the same curve when viewed as a curve over the $y$-axis, which is the same thing as viewing the graph of $y\mapsto x:={\root n\of{\mathstrut y}}$, gets ever flatter as $y$ increases. This implies that for given $\delta>0$ and $y\geq\delta$ one has $${\root n\of {y+\delta}}-{\root n\of {\mathstrut y}}\leq {\root n\of {\mathstrut y}}-{\root n\of {y-\delta}}\ .$$

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.